JP2019014407A - Runway recognition device - Google Patents

Abstract

To provide a technology that is able to accurately recognize a shape of a distant runway even in the case where a shape of a runway near a vehicle cannot be recognized.SOLUTION: An image processing system comprises a noise reduction unit, a moving blur reduction unit, and a similar image specifying unit. Using a noise reduction image and a moving blur noise reduction image obtained by reducing moving blur with respect to the noise reduction image obtained by reducing noise of a captured image input to the image processing system, the similar image specifying unit specifies a similar image, which is similar to an image in which the moving blur is added to the moving blur noise reduction image and similar to the noise reduction image and in which a rough surface is restricted to a degree higher than a predetermined degree. By reducing the moving blur from the similar image, the moving blur reduction unit obtains a processed image in which the moving blur and the noise are reduced from the captured image.SELECTED DRAWING: Figure 1

Description

  The present disclosure relates to a technique for recognizing a running path of a vehicle.

  For automatic driving of a vehicle, it is required to recognize the traveling lane of the host vehicle, for example, the traveling lane of the host vehicle if it is a road. The recognition of the runway is executed based on the white line, for example, by detecting a white line that is a marking line of the runway in a captured image obtained by an imaging device such as a camera. When the traveling path of the host vehicle is recognized, operations necessary for automatic driving such as obstacle detection on the traveling path, automatic brake operation, and estimation of lane departure can be performed. Therefore, in order to realize more stable automatic driving, a technique for accurately recognizing the shape of the runway from the vicinity of the current position to the distance is desired. Generally, the accuracy of recognition of a distant track shape is lower than that of a nearby track shape. Therefore, in the captured image, first, a white line in the nearby runway is detected to recognize the shape of the nearby runway, and then in the far area (generally, the upper area of the captured image) in the captured image, there is a white line in the distance. A method has been proposed in which narrowing is performed based on the shape of a nearby road that recognizes an expected area, a white line is detected in the narrowed area, and a far road shape is recognized based on the detected white line (see Patent Document 1). ).

JP 2013-196341 A

  In the method of Patent Document 1, when there is no white line in a region in the vicinity of the host vehicle such as when the host vehicle is traveling at an intersection, the nearby runway shape cannot be recognized. For this reason, there is a problem in that it is not possible to narrow down the region where the white line is expected to exist in the far region, and it is impossible to accurately recognize the distant road shape. Therefore, there is a demand for a technology that can accurately recognize a distant road shape even when the shape of the road near the vehicle cannot be recognized.

  This indication is made in order to solve at least one part of the above-mentioned subject, and can be realized as the following forms.

  According to one form of the present disclosure, a road recognition device (100, 100a to 100k) that is mounted on a vehicle and recognizes the road of the vehicle is provided. The runway recognition apparatus includes: a neighborhood shape calculation unit (101) that repeatedly calculates a neighborhood runway shape that is a runway shape in the vicinity of the vehicle based on captured images repeatedly obtained by the imaging device (21) mounted on the vehicle; A processing area (Er1 to Er4, El1 to El4) in each captured image is set on the basis of the calculated nearby running path shape, and a far running path shape that is a far running path shape of the vehicle in the processing area is repeatedly calculated. A distant shape calculation unit (102); a vicinity shape storage unit (103) for storing the calculated near-runway shape as a history in a time series; and a calculated far-runway shape as a history in a time series A remote shape storage unit (104); and the remote shape calculation unit, when the near road shape is not calculated, the history of the near road shape, A history of rectangular track shape, using at least one of, sets the processing region.

  According to the runway recognition apparatus of this embodiment, when the nearby runway shape cannot be calculated using the captured image, the processing region is determined using at least one of the near runway shape history and the far runway shape history. Since it is set, it is possible to accurately recognize a distant road shape even when the road shape near the vehicle cannot be recognized.

  The present disclosure can also be realized in various forms other than the road recognition device. For example, the present invention can be realized in the form of a vehicle including a road recognition device, a road recognition method, a computer program for recognizing the road, a storage medium storing the computer program, and the like.

The block diagram which shows schematic structure of the runway recognition apparatus as one Embodiment of the image processing apparatus of this indication. The flowchart which shows the procedure of the near runway shape calculation process in 1st Embodiment. Explanatory drawing which shows an example of a captured image. The flowchart which shows the procedure of the far road shape calculation process in 1st Embodiment. Explanatory drawing which expands and shows an example of a distant track area. Explanatory drawing which shows an example of a captured image. The flowchart which shows the procedure of the target passage point setting process in 1st Embodiment. The flowchart which shows the procedure of the principal part of the far road shape calculation process in 2nd Embodiment. The flowchart which shows the procedure of the target passage point setting process in 3rd Embodiment. The block diagram which shows schematic structure of the runway recognition apparatus of 4th Embodiment. The flowchart which shows the procedure of the principal part of the far road shape calculation process in 4th Embodiment. The flowchart which shows the procedure of the target passage point setting process in 5th Embodiment. The flowchart which shows the procedure of the principal part of the far road shape calculation process in 6th Embodiment. Explanatory drawing which shows an example of the setting content of the evaluation value table in 6th Embodiment. Explanatory drawing which shows an example of the setting content of the integrated likelihood table in 6th Embodiment. The block diagram which shows schematic structure of the runway recognition apparatus of 7th Embodiment. The flowchart which shows the procedure of the execution determination process in 7th Embodiment. The block diagram which shows schematic structure of the runway recognition apparatus of 8th Embodiment. Explanatory drawing which shows an example of the process area | region set in 8th Embodiment. The block diagram which shows schematic structure of the runway recognition apparatus in 9th Embodiment. A block diagram showing a schematic structure of a runway recognition device in a 10th embodiment. A block diagram showing a schematic structure of a runway recognition device in an 11th embodiment. The flowchart which shows the procedure of the principal part of the far road shape calculation process in 11th Embodiment. The block diagram which shows schematic structure of the runway recognition apparatus in 12th Embodiment. The flowchart which shows the procedure of the principal part of the far road shape calculation process in 12th Embodiment. Explanatory drawing which shows typically the measurement distance to a pole when there is no attitude change of the own vehicle in 12th Embodiment. Explanatory drawing which shows typically a mode that the distance to a pole when nose-down arises in 12th Embodiment based on a captured image. Explanatory drawing which shows typically a mode that the distance to a pole when nose-down arises in 12th Embodiment based on the measurement result of a millimeter wave radar. A block diagram showing a schematic structure of a runway recognition device in a 13th embodiment. The flowchart which shows the procedure of the pattern matching process in 14th Embodiment. A block diagram showing a schematic structure of a runway recognition device in a 15th embodiment. The flowchart which shows the procedure of the principal part of the far road shape calculation process in 15th Embodiment. Explanatory drawing which shows an example of the process area | region set in 15th Embodiment. A block diagram showing a schematic structure of a runway recognition device in a 16th embodiment. The flowchart which shows the procedure of the principal part of the far road shape calculation process in 16th Embodiment. Explanatory drawing which shows the white line candidate determined by step S805, and its score value. A block diagram showing a schematic structure of a runway recognition device in a 17th embodiment. The flowchart which shows the procedure of the principal part of the far road shape calculation process in 17th Embodiment. Explanatory drawing which shows the example of determination of the normality of the white line candidate in 17th Embodiment. The flowchart which shows the procedure of the process which determines a white line candidate in 18th Embodiment.

A. First embodiment:
A1. Device configuration:
A travel path recognition apparatus 100 of the present embodiment shown in FIG. 1 is mounted on a vehicle (not shown) and recognizes the travel path of the vehicle. In the present embodiment, the vehicle on which the travel path recognition device 100 is mounted is also referred to as “own vehicle”. The own vehicle is a vehicle capable of automatic driving. The own vehicle may be either a vehicle that switches between automatic driving and manual driving (driving by the driver) and a vehicle that executes only automatic driving of these two types of driving. The travel path recognition apparatus 100 is configured by an ECU (Electronic Control Unit) 10 including a CPU and a memory (not shown). Similarly, the travel path recognition device 100 is electrically connected to an imaging unit 21, a vehicle speed sensor 22, a yaw rate sensor 23, and a vehicle control unit 31 that are mounted on the vehicle.

  The imaging unit 21 includes two cameras (so-called stereo cameras) arranged in the horizontal direction. Each camera includes a condensing lens and a light receiving element, and images the front of the host vehicle to obtain a captured image. When the ignition of the host vehicle is turned on, the imaging unit 21 repeatedly executes captured image acquisition. The vehicle speed sensor 22 measures the moving speed of the host vehicle. The vehicle speed sensor 22 repeatedly detects the vehicle speed when the ignition of the host vehicle is turned on. The yaw rate sensor 23 detects the yaw rate (rotational angular velocity) of the host vehicle. The yaw rate sensor 23 repeatedly executes yaw rate detection when the ignition of the host vehicle is turned on. The vehicle control unit 31 controls the movement of the vehicle. Examples of the vehicle control unit 31 include a plurality of ECUs such as an ECU that controls the operation of a drive mechanism such as an internal combustion engine or an electric motor, an ECU that controls an electronically controlled brake system (ECB), and an ECU that controls the steering angle of a wheel. Consists of.

  When the CPU included in the travel path recognition apparatus 100 executes a control program stored in the memory, the travel path recognition apparatus 100 includes a near shape calculation unit 101, a far shape calculation unit 102, a near shape storage unit 103, and a far shape storage unit. 104 and the target passing point setting unit 105.

  The neighborhood shape calculation unit 101 executes a neighborhood track shape specifying process that will be described later, and based on the captured images repeatedly obtained by the imaging unit 21, the shape of the track that the vehicle travels and is near the host vehicle. (Hereinafter referred to as “neighboring runway shape”) is repeatedly calculated. “Near the host vehicle” means a region within a predetermined distance from the host vehicle in the region ahead of the imaging unit 21 in the traveling direction. In the present embodiment, the above-mentioned predetermined distance that defines the vicinity of the host vehicle is 7 m (meters). The distance is not limited to 7 m and may be an arbitrary distance. In this embodiment, “the shape of the runway” is defined by the width of the runway, the positions of both ends in the width direction of the runway, and the curvature (curvature radius) of the runway. Therefore, “calculate the shape of the nearby runway” means the runway on which the host vehicle travels, the runway width in the vicinity of the host vehicle (hereinafter referred to as “neighboring runway”), and both ends in the width direction of the runway. And the curvature (curvature radius) of the runway are calculated respectively.

  The distant shape calculation unit 102 repeatedly calculates the shape of the distant road (hereinafter referred to as “distant road shape”) that is the travel path of the host vehicle by executing the far road shape specifying process described later. To do. Although the detailed procedure will be described later, the outline shape calculation unit 102 is a target of processing for calculating the distance road shape in the captured image based on the vicinity road shape calculated by the vicinity shape calculation unit 101. An area (hereinafter simply referred to as a “processing area”) is specified, and a far road shape is calculated within the processing area. “Calculate the shape of the far road” refers to the width of the far road (hereinafter referred to as “far road”), the positions of both ends in the width direction of the far road, This means calculating the curvature (curvature radius) of the runway.

  The neighborhood shape storage unit 103 stores the neighborhood track shape calculated by the neighborhood shape calculation unit 101 as a history in time series. Specifically, each parameter indicating the calculated nearby runway shape is stored together with the calculation time. The far shape storage unit 104 stores the far road shape calculated by the far shape calculation unit 102 as a history in time series. Specifically, each parameter indicating the calculated far road shape is stored together with the calculation time.

  The target passing point setting unit 105 sets a passing point targeted by the host vehicle by executing a target passing point setting process described later. This target passing point is used as a target passing point when the host vehicle automatically operates. Basically, the target passing point is set based on the near road shape, but is exceptionally set based on the far road shape when the calculation accuracy of the near road shape is low. This is because the calculation accuracy of the near running road shape is basically higher than the calculation accuracy of the far running road shape.

A2. Neighborhood road shape calculation processing:
2 is periodically and repeatedly executed when the ignition of the host vehicle is turned on. The neighborhood shape calculation unit 101 extracts the edge of the nearby runway in the obtained captured image, more specifically, the edge of the lane marking that divides the neighborhood runway (step S105).

  The neighborhood shape calculation unit 101 detects a straight line by performing a well-known Hough transform on the extracted edge (step S110), and calculates a straight line of a partition line (white line) candidate from the detected straight line (step S110). S115). Specifically, among the detected straight lines, those having a large number of votes for Hough transform are calculated as white line candidates.

  As shown in FIG. 3, the captured image F1 includes a lane Ln1 and an opposite lane Ln2 that are the traveling path of the host vehicle. An area including a far road (hereinafter referred to as a “far road area”) Af1 located on the upper side of the captured image F1 and an area including a nearby road located on the lower side of the captured image F1 (hereinafter referred to as “near road) The edge of the lane lines Lr1 and Ll1 of the lane Ln1 is extracted in the nearby track area An1, and the straight lines representing the lane lines Lr1 and Ll1 are calculated as white line candidates.

  As shown in FIG. 2, the neighborhood shape calculation unit 101 narrows down white line candidates (step S120). Specifically, for example, the ratio of the contrast of the white line candidate to the road surface around the white line candidate in the photographed image is higher than a predetermined threshold, or the difference between the luminance of the white line candidate part and the surrounding luminance is a predetermined threshold. Limit white line candidates to those larger than the above. In addition, it may be narrowed down in consideration of various characteristics such as line thickness and total extension distance. Then, one white line candidate closest to the left-right direction from the center of the vehicle is selected.

  The neighboring shape calculation unit 101 determines whether or not the white line candidates are narrowed down in step S120 (step S125). If it is determined that the white line candidates are not narrowed down, that is, there is no white line candidate (step S125: NO), The process returns to step S105. On the other hand, when it is determined that there is a white line candidate (step S125: YES), the neighboring shape calculation unit 101 converts the edge points constituting the white line candidate narrowed down in S120 into plane coordinates (S130). ). The plane coordinates are coordinates when the road surface is assumed to be a plane. Here, the coordinate conversion of the edge point on the captured image is performed based on the position / orientation information (hereinafter, referred to as “imaging parameter”) of the camera constituting the imaging apparatus obtained in advance by calibration. By using the edge point as information on the plane coordinates in this way, it becomes possible to easily combine with the coordinate information of the edge point based on the past captured image.

  The neighborhood shape calculation unit 101 calculates the neighborhood track shape from the plane coordinates calculated in step S130 (step S135). As described above, the near runway shape means the width of the runway, the positions of both ends in the width direction of the runway, and the curvature (curvature radius) of the runway. As a calculation method of each of these parameters, for example, a well-known method such as a method described in JP2013-196341A may be adopted.

  The neighborhood shape calculation unit 101 outputs the neighborhood road shape calculated in step S135 to the far shape calculation unit 102 and the target passing point setting unit 105, and stores it in the neighborhood shape storage unit 103 together with the calculated time (step). S140).

A3. Distant road shape calculation processing:
The near road shape calculation process shown in FIG. 4 is repeatedly executed periodically when the ignition of the host vehicle is turned on. The distant shape calculation unit 102 determines whether or not there is an input of the near road shape at a predetermined time (step S205). As described above, when step S140 of the nearby running road shape calculation process is executed, it is determined that “input is present”. On the other hand, if it is determined that there are no white line candidates as a result of narrowing down the white line candidates in step S125 of the nearby runway shape calculation process, step S140 is not executed, and the nearby runway shape is calculated as a far-end shape at a predetermined time. It is not input to the part 102.

  When it is determined that there is an input of the near track shape (step S205: YES), the far shape calculation unit 102 extends the near track (division line) by using the input near track shape, and the far track shape. Is estimated (step S210). The far shape calculation unit 102 sets a processing region based on the estimated far road shape (step S215). The processing area means an area narrowed down as a target area from which edge points of lane markings are extracted when calculating (specifying) a far road shape among far road areas. In other words, it means a region to be processed in which the edge point of the lane marking is extracted and the far road shape is calculated.

  In FIG. 5, the far road area Af1 in the captured image F1 of FIG. 3 is schematically illustrated in an enlarged manner. In FIG. 5, two processing areas Er1 and El1 are set in the far road area Af1 based on the far road shape estimated in step S210. FIG. 3 also shows the processing areas Er1 and El1. Hereinafter, a method for setting these two processing areas Er1 and El1 will be described. On the lane marking Lr1, a point Pr11 that is a distance from the nearest point Pr1 and the vanishing point P0 by a predetermined distance d1 is determined. The vanishing point P0 is obtained as a point where the two partition lines Lr1 and Ll1 intersect. Next, with respect to the two determined points Pr1 and Pr11, two sides (short sides) having a predetermined distance in the horizontal direction are obtained around the corresponding points, and the ends of the sides are defined as a dividing line. Two sides (long sides) are defined by connecting along Lr1. A region surrounded by the two short sides and the two long sides thus determined is set as a processing region Er1. Similarly, on the division line L11, a point P11 that is the nearest side from the point P11 and the vanishing point P0 that is a predetermined distance d1 is determined. Next, with respect to the two determined points Pl1 and Pl11, two sides (short sides) having a predetermined distance in the horizontal direction are obtained with the corresponding point as the center, and the ends of the sides are separated from each other. Two sides (long sides) are defined by connecting along Ll1. A region surrounded by the two short sides and the two long sides thus determined is set as a processing region El1. As shown in FIGS. 3 and 5, when the running road (lane Ln1) is a straight line and the dividing lines Lr1 and Ll1 on both sides are straight lines, the shapes of the processing regions Er1 and El1 are both captured images. It becomes a trapezoid shape.

  As shown in FIG. 4, after setting the processing area (after step S215), the far shape calculation unit 102 extracts an edge that divides the far road in the processing area (step S220). The Hough transform (step S225), white line candidate calculation (step S230), candidate narrowing down (step S235), determination of presence / absence of candidates (step S240), and edge point plane conversion (step S245) executed thereafter are all described above. Since this is the same as steps S110 to S130 of the nearby runway shape calculation process, detailed description thereof is omitted. In addition, when it determines with there being no white line candidate in step S240 (step S240: NO), a far road shape calculation process is complete | finished.

  After execution of step S245, the distant shape calculation unit 102 calculates the distant traveling road shape from the plane coordinates calculated in step S240 (step S250). Since this calculation method is the same as that of step S135 of the above-mentioned neighborhood track shape calculation process, detailed description thereof is omitted.

  The far shape calculation unit 102 outputs the far road shape calculated in step S250 to the target passing point setting unit 105, and stores it in the far shape storage unit 104 together with the calculated time (step S255).

  In step S205 described above, when it is determined that there is no input of a nearby runway shape at a predetermined time (step S205: NO), the far road shape calculation unit 102 stores the far roadway shape calculated at the previous timing as a far shape memory. It is determined whether it exists in the part 104 (step S270).

  The captured image F2 illustrated in FIG. 6 is an example of an image obtained when the host vehicle is stopped before the intersection CR. In the captured image F2, although the lane markings Lr2 and Ll2 are slightly shown at the lower end (front end) of the adjacent runway region An2, the pedestrian crossing Pd1 shown below the nearby runway region An2 is ahead. The lane marking is not shown. Therefore, in this case, the white line candidates are not narrowed down, and the near road shape is not input to the far shape calculation unit 102, and it is determined in step S205 described above that there is no input of the near road shape at a predetermined time. In the captured image F2, the distant runway region Af2 includes a pedestrian crossing Pd2 that exists across the intersection CR and a continuation of the lane markings Lr2 and Ll2.

  As illustrated in FIG. 4, when it is determined that the far road shape calculated at the previous timing is not in the far shape storage unit 104 (step S <b> 270: NO), the far road shape calculation process ends. On the other hand, if it is determined that there is a history of the far road shape (step S270: YES), the far shape calculation unit 102 stores the previous far road shape stored in the far shape storage unit 104, the imaging parameters, and the like. Based on the information indicating the movement of the vehicle, a processing area is set in the far road area (step S275). The information indicating the movement of the vehicle means, for example, a moving speed of the host vehicle obtained from the vehicle speed sensor 22, a yaw rate obtained from the yaw rate sensor 23, a steering angle obtained from the vehicle control unit 31, and the like. In step S275, the far shape calculation unit 102 corrects the far road shape calculated last time from the imaging parameters and the motion of the vehicle, and estimates the current far road shape (in the current captured image). Then, the lane marking in the far lane area is specified from the estimated far lane shape, and the processing area is set based on the lane marking. Since the method for setting the processing area based on the lane marking is the same as the method described for step S215 described above, detailed description thereof is omitted. In the example of FIG. 6, two processing areas Er2 and El2 are set in the far road area Af2. After execution of step S275, the above-described steps S220 to S255 are executed.

A4. Target passing point setting process:
The target passing point setting process shown in FIG. 7 is repeatedly executed periodically when the ignition of the host vehicle is turned on. The target passing point setting unit 105 determines whether or not there is an input of a nearby runway shape at a predetermined time (step S305). As described above, the near road shape calculation process is periodically executed, and when step S140 is executed, the near road shape is periodically input to the target passing point setting unit 105. On the other hand, if the nearby runway shape is not calculated, the nearby runway shape is not input to the target passing point setting unit 105 at a predetermined time.

  When it is determined that there is an input of a nearby runway shape (step S305: YES), the target passing point setting unit 105 sets a target passing point based on the nearby runway shape (step S310). Specifically, the target passing point setting unit 105 sets the center point of the neighboring runway in the neighboring runway region as the target passing point. For example, in the example of FIG. 3, the center point PT1 in the horizontal direction of the lane markings Lr1 and Ll1 at both ends of the lane Ln1 and the center in the depth direction in the adjacent track area An1 is set as the target passing point.

  As illustrated in FIG. 7, when it is determined in step S305 described above that there is no input of the near road shape (step S305: NO), the target passing point setting unit 105 determines whether there is an input of the far road shape. Is determined (step S315). When step S <b> 255 of the far road shape calculation process described above is executed, the far road shape is input to the target passing point setting unit 105. On the other hand, when step S255 is not executed, that is, when it is determined that there is no input of the near track shape (step S205: NO) and there is no far track shape calculated at the previous timing (step S270). : NO), the far road shape is not input to the target passing point setting unit 105.

  When it is determined that there is an input of the far road shape (step S315: YES), the target passing point setting unit 105 sets a target passing point based on the far road shape (step S330). Specifically, the target passing point setting unit 105 sets a point closest to the far side in the horizontal direction of the far road in the far road area as the target passing point. For example, in the example of FIG. 6, the point PT2 closest to the center in the horizontal direction of the division lines Lr2 and Ll2 and in the far travel area Af2 is set as the target passing point.

  After execution of step S310 described above, after execution of step S330, or when it is determined in step S315 that there is no far road shape input (step S315: NO), the target passing point setting process ends.

  According to the runway recognition device 100 of the first embodiment described above, when the neighborhood runway shape cannot be calculated using the captured image, the processing region is set using the previous far runway shape, so that the vehicle Even if the runway shape in the vicinity of can not be recognized, the far runway shape can be recognized with high accuracy. In addition, when the near runway shape is output, the target pass point is set based on the near runway shape, so that the target pass point can be set with high accuracy. In addition, since the target passing point is set based on the far traveling road shape when the nearby running road shape is not output, it can be avoided that the target passing point cannot be set even when the neighboring traveling road shape is not output, Even in such a case, the target passing point can be set based on the recognized road shape. For this reason, the target passing point can be set with high accuracy.

B. Second embodiment:
Since the device of the road recognition device of the second embodiment is the same as the device configuration of the road recognition device 100 of the first embodiment, the same components are denoted by the same reference numerals, and detailed description thereof is omitted. . The runway shape calculation process of the second embodiment is different from the runway shape calculation process of the first embodiment shown in FIG. 4 in that step S209 is added and executed as shown in FIG. It is.

  In step S205, when it is determined that there is no input of the near track shape (step S205: YES), the far shape calculation unit 102 does not continuously input the near track shape beyond a predetermined threshold time. It is determined whether or not (step S209). In this embodiment, 1 second is set as a predetermined period (predetermined time). In addition, you may set not only 1 second but the arbitrary time longer than the acquisition interval of a captured image. When it is determined that the input of the adjacent runway shape does not continue beyond the predetermined threshold (step S209: YES), the far runway shape calculation process ends. On the other hand, when it is determined that there is no continuous input of the adjacent track shape exceeding the predetermined threshold (that is, there is an input) (step S209: NO), the above-described step S270 is performed. Executed. As described above, when the input of the adjacent runway shape is not continued beyond the predetermined threshold time, the far runway shape calculation process is terminated by calculating the far runway shape using the history of the far runway shape. This is to prevent the calculation accuracy of the distant traveling road shape from being lowered due to continuing for a long time. Note that the threshold time in step S209 corresponds to a subordinate concept of the first period in the claims.

  The travel path recognition apparatus of the second embodiment described above has the same effect as the travel path recognition apparatus 100 of the first embodiment. In addition, since the calculation of the far road shape is stopped when the near road shape is not continuously calculated beyond a predetermined period, it is possible to suppress a decrease in the calculation accuracy of the far road shape. For this reason, it can suppress that the far road shape calculated with low precision is memorize | stored as a log | history, and can suppress that the specific precision of a process area falls.

C. Third embodiment:
Since the device of the road recognition device of the third embodiment is the same as the device configuration of the road recognition device 100 of the first embodiment, the same components are denoted by the same reference numerals, and detailed description thereof is omitted. . The target passing point setting process of the third embodiment is different from the target passing point setting process of the first embodiment shown in FIG. 7 in that the steps S319 and S320 are added and executed as shown in FIG. The procedure is the same.

  As shown in FIG. 9, when it is determined in step S315 that there is no far road shape input (step S315: NO), the target passing point setting unit 105 sets a predetermined threshold time (predetermined threshold time). It is determined whether or not there is no input of the far road shape (step S319). In the present embodiment, the above-described predetermined threshold time (predetermined threshold time) is set to 2 seconds. When it is determined that there is no far road shape input exceeding the predetermined threshold time (step S319: YES), the target passing point setting process ends. On the other hand, when it is determined that there is no input of the far road shape exceeding the predetermined threshold time (that is, there is an input within the threshold time) (step S319: NO), the target passing point The setting unit 105 calculates the current far road shape based on the far road shape history stored in the far shape storage unit 104 (step S320). In this embodiment, this step S320 is executed by replacing the “previous far road shape” with the “latest far road shape” in step S275 of the far road shape calculation process in the first embodiment shown in FIG. It means the process of executing steps S215 to S255. In addition, it is not limited to the “latest far-off road shape”, and for example, these far-runway shapes are averaged using a predetermined number of far-off road shapes counted from the newer side, and then the process of step S275 is performed. S220 to S225 may be executed. That is, the current far road shape may be calculated by an arbitrary method using the history of the far road shape. Further, instead of the history of the far road shape, or in addition to the history of the far road shape, the current far road shape may be calculated using the history of the near road shape. For example, a far road shape may be obtained by simply extending (extrapolating) a lane line specified by the latest neighboring road shape, and the far road shape may be used as the current far road shape. Alternatively, the step S210 is executed by replacing the “(input) near road shape” in step S210 of the far road shape calculation process shown in FIG. 4 with the “latest near road shape”, and then the steps S220 to S255 are executed. Then, the current far road shape may be calculated. Furthermore, the average value of the far road shape obtained by simply extending (extrapolating) the lane line specified by the latest neighboring road shape and the past far road shape is obtained, and the average value is shown in FIG. It is also possible to execute step S275 in place of the “previous far road shape” in step S275 of the far road shape calculation process, and then execute steps S220 to S255 to calculate the current far road shape. That is, in general, any method for calculating the current far road shape using at least one of the history of the near road shape and the history of the far road shape may be executed in step S320. After step S320 is executed, step S330 described above is executed. The threshold time in step S320 corresponds to a subordinate concept of the second period in the claims.

  The travel path recognition apparatus of the third embodiment described above has the same effect as the travel path recognition apparatus 100 of the first embodiment. In addition, if the far road shape is not input beyond the threshold time, that is, if the far road shape is not calculated beyond the threshold time, at least one of the history of the near road shape and the history of the far road shape Since this far-running road shape is calculated using, it is possible to prevent the far-running road shape from being frequently calculated. Moreover, it can suppress that the quantity of the log | history of a distant track shape decreases, and the specific precision of a process area | region falls. On the other hand, when the period during which the far road shape is not calculated exceeds the threshold time, the calculation of the far road shape is stopped, so that the far road shape recognized with low accuracy can be suppressed from being stored as a history.

D. Fourth embodiment:
The road recognition device 100a of the fourth embodiment shown in FIG. 10 is different from the road recognition device 100 of the first embodiment in that it functions as the vehicle position determination unit 106. Since the other structure of the track recognition apparatus 100a of 4th Embodiment is the same as the track recognition apparatus 100 of 1st Embodiment, the detailed description is abbreviate | omitted.

  The vehicle position determination unit 106 determines whether or not the current position of the host vehicle is within the intersection area. The intersection area means an area including an intersection and a neighborhood area of the intersection. The neighborhood area of the intersection means an area within a predetermined distance from the center of the intersection, and in this embodiment, means an area within 50 m from the center of the intersection. In addition, you may mean the area | region within the arbitrary distance recognized as being not only 50m but the intersection vicinity. In the present embodiment, the vehicle position determination unit 106 determines whether or not the current position is within the intersection area based on whether or not the traffic light is shown in the captured image and the size is equal to or greater than a predetermined size. To do. When the traffic light is shown to be larger than a predetermined size, the position of the own vehicle is located in or near the intersection. The determination may be made based on the position information of the own vehicle obtained by a navigation system (not shown) mounted on the own vehicle.

  As shown in FIG. 11, the far road shape calculation process of the fourth embodiment is different from the road shape calculation process of the second embodiment in that steps S206, S207, and S208 are added, and the other procedures are as follows. Are the same.

  In step S205, when it is determined that there is no input of a nearby runway shape (step S205: YES), the far shape calculation unit 102 determines whether or not the current position of the host vehicle is within the intersection area (step S206). ).

  When it is determined that the current position of the host vehicle is not within the intersection area (step S206: NO), the far shape calculation unit 102 sets the threshold time in step S209 described above to a normal time (step S207). In this embodiment, 1 second is set as the normal time. In addition, you may set not only 1 second but the arbitrary time longer than the acquisition interval of a captured image.

  On the other hand, when it is determined that the current position of the host vehicle is within the intersection area (step S206: YES), the far-field shape calculation unit 102 sets the predetermined threshold time in step S209 described above from the normal time. Is set to a longer time (step S208). In this embodiment, 2 seconds is set as a longer time. The time is not limited to 2 seconds, and may be set to an arbitrary time longer than the normal time.

  After execution of step S207 or S208, the above-described step S209 is executed. That is, the distant shape calculation unit 102 determines whether or not the input of the nearby runway shape is continued beyond a predetermined threshold time. In the fourth embodiment, the threshold time in step S209 is set in step S207 or S208. As described above, when the current position of the host vehicle is in the intersection area, the threshold time in step S209, that is, a reference for determining whether to stop calculating the far road shape, The reason why the time during which continuous input is not set is set to be long is as follows. That is, when the current position of the host vehicle is within the intersection area, it is clear that the calculation of the nearby track shape will resume after a short period of time as the host vehicle travels, so the calculation of the far track shape is not stopped. However, this is because there is little influence on the calculation of the shape of the far road. Moreover, by doing in this way, it is possible to suppress the amount of the history of the far road shape from being unnecessarily reduced due to the stop calculation of the far road shape. On the other hand, if the current position of the host vehicle is not within the intersection area, if the neighboring runway shape is not continuously input, there is some problem, for example, a malfunction of the captured image or a fault of the runway recognition device 100. It may have occurred. In such a case, if calculation of the far road shape is continued, automatic driving may be performed based on the far road shape with low accuracy. In this embodiment, the threshold time in step S209 in this case is set as the normal time. , Do not set long.

  The runway recognition apparatus of the fourth embodiment described above has the same effect as the runway recognition apparatus 100 of the first embodiment and the second embodiment. In addition, since the length of the threshold time in step S209 is set according to whether or not the current position of the host vehicle is in the intersection area, the reason why the neighboring runway shape is not calculated is that "the host vehicle is in the intersection area. If it is clear that the calculation of the nearby runway shape will resume after a short period of time as the vehicle travels, the calculation of the far runway shape is suppressed from being stopped, It is possible to suppress the amount of the history of the far road shape from being inadvertently reduced. In addition, when the current position of the host vehicle is not in the intersection area, the threshold time is set to a shorter time than in the intersection area, so that it is possible to suppress erroneous recognition of the far road shape.

E. Fifth embodiment:
Since the device of the road recognition device of the fifth embodiment is the same as the device configuration of the road recognition device 100 of the first embodiment, the same components are denoted by the same reference numerals, and detailed description thereof is omitted. . The target passing point setting process of the fifth embodiment is different from the target passing point setting process of the third embodiment in that steps S316, S317, and S318 are added and executed as shown in FIG. Are the same.

  As shown in FIG. 12, when it is determined in step S315 that there is no input of the far road shape (step S315: NO), the target passing point setting unit 105 determines whether or not the current position of the host vehicle is within the intersection area. Is determined (step S316). Since step S316 is the same process as step S206 of the fourth embodiment described above, a detailed description of the procedure is omitted.

  When it is determined that the current position of the host vehicle is not within the intersection area (step S316: NO), the target passing point setting unit 105 sets the threshold time in step S319 described above to a normal time (step S317). In this embodiment, 1 second is set as the normal time. In addition, you may set not only 1 second but the arbitrary time longer than the acquisition interval of a captured image.

  On the other hand, when it is determined that the current position of the host vehicle is in the intersection area (step S316: YES), the target passing point setting unit 105 sets the predetermined threshold time in step S319 described above to a normal time. A longer time is set (step S318). In this embodiment, 2 seconds is set as a longer time. The time is not limited to 2 seconds, and may be set to an arbitrary time longer than the normal time.

  After execution of step S317 or S318, the above-described step S319 is executed. That is, the distant shape calculation unit 102 determines whether or not the distant running road shape input continues beyond a predetermined threshold time. In the fifth embodiment, the threshold time in step S319 is set in step S317 or S318. As described above, when the current position of the host vehicle is within the intersection area, the threshold time in step S319, that is, a reference for determining whether to stop the calculation (estimation) of the far road shape is a distant place The reason why the time during which the far road shape as the result of the running road shape calculation process is not input is set longer is the same as the reason why the threshold time is set longer in step S209 in the fourth embodiment described above.

  The travel path recognition apparatus in the fifth embodiment described above has the same effect as the travel path recognition apparatus 100 of the first embodiment and the third embodiment. In addition, since the length of the threshold time in step S319 is set according to whether or not the current position of the host vehicle is within the intersection area, the reason why the far road shape is not calculated by the far road shape calculation process is If it is clear that the vehicle is present in the intersection area and the calculation of the far road shape is resumed after a short period of time as the vehicle travels, the calculation of the far road shape is stopped. This can be suppressed, and the amount of the history of the far road shape can be prevented from being inadvertently reduced. In addition, when the current position of the host vehicle is not in the intersection area, the threshold time is set to a shorter time than in the intersection area, so that it is possible to suppress erroneous recognition of the far road shape.

F. Sixth embodiment:
Since the device of the road recognition device of the sixth embodiment is the same as the device configuration of the road recognition device 100 of the first embodiment, the same components are denoted by the same reference numerals, and detailed description thereof is omitted. . The runway shape calculation process of the sixth embodiment is different from the runway shape calculation process of the first embodiment in that steps S405 and S410 are added and executed as shown in FIG. .

  As shown in FIG. 13, when it is determined in step S205 that there is an input of a nearby runway shape at a predetermined time (step S205: YES), the far-field shape calculation unit 102 is sure of the input nearby runway shape. Is evaluated (step S405). Specifically, the distant shape calculation unit 102 specifies a distance along the traveling direction of the nearby traveling path (hereinafter referred to as “neighboring traveling distance”) from the nearby traveling path shape input in step S205, and the nearby traveling path. An evaluation value is determined with reference to the evaluation value table based on the distance. The evaluation value table is stored in advance in a memory (not shown) provided in the travel path recognition apparatus 100.

  As shown in FIG. 14, in the evaluation value table, a nearby running distance and an evaluation value are set in association with each other. In FIG. 14, the horizontal axis indicates the near road distance, and the vertical axis indicates the evaluation value. As shown in FIG. 14, in the evaluation value table, the evaluation value is set to be higher as the nearby running distance is longer. This is because the calculation accuracy of the neighborhood runway shape becomes higher as the neighborhood runway distance is longer. In the example shown in FIG. 14, the rate of change of the evaluation value with respect to the change in the vicinity road distance when the vicinity road distance is 0 to the distance La is the largest, and the evaluation value with respect to the change in the vicinity road distance from the distance La to the distance Lb is The change between the adjacent runway distance and the evaluation value is set in three stages so that the change rate of the evaluation value with respect to the change in the adjacent runway distance when the change rate is the second largest and larger than the distance Lb is the lowest. Yes. In addition, you may set in steps not only in three steps but in arbitrary steps. Moreover, you may change continuously exponentially.

  As shown in FIG. 13, the far-distance calculation unit 102 determines the near-run road shape calculated this time, that is, the near-run road shape input this time, the far-run road shape calculated last time, according to the determined evaluation value. Based on the imaging parameters and the information indicating the movement of the vehicle, a processing area is set (step S410). Hereinafter, the specific procedure of step S410 will be described.

  First, the distant shape calculation unit 102 is based on the evaluation value determined in step S405 and the degree of likelihood of the processing area estimated based on the near-run road shape input this time (hereinafter referred to as “likelihood”). Then, the likelihood of the processing area estimated based on the previous far road shape stored in the far shape storage unit 104 is specified with reference to the integrated likelihood table. In the present embodiment, the “processing area estimated based on the nearby runway shape input this time” is also referred to as a “first estimation processing area”. In addition, the “processing area estimated based on the previous far road shape” is also referred to as a “second estimation processing area”.

  The integrated likelihood table shown in FIG. 15 is stored in advance in a memory (not shown) included in the travel path recognition apparatus 100. As shown in FIG. 15, in the integrated likelihood table, the likelihood of the first estimation processing region and the likelihood of the second estimation processing region are set in association with the evaluation value. In FIG. 15, the horizontal axis indicates the evaluation value, and the vertical axis indicates the likelihood of the first estimation processing region and the likelihood of the second estimation processing region. In FIG. 15, an area A1 below the straight line AL indicates the likelihood of the first estimation processing area, and an area AL2 (upper limit 100%) above the straight line AL indicates the likelihood of the second estimation processing area. In this embodiment, the sum of the likelihood of the first estimation processing region and the second estimation processing region likelihood is normalized so as to be always 100%. In the integrated likelihood table, the higher the evaluation value, that is, the higher the likelihood of the input nearby running road shape, the higher the likelihood of the first estimation processing region and the lower the likelihood of the second estimation processing region. For example, as shown in FIG. 15, when the evaluation value is a value E1, the likelihood of the first estimation processing region is 70% (0.7), and the likelihood of the second estimation processing region is 30%. (0.3).

  Next, the distant shape calculation unit 102 estimates a processing region based on the current near road shape. That is, the first estimation processing area is set. In this embodiment, this estimation is realized by executing steps S210 and S215 of the far road shape calculation process in the first embodiment. Next, the distant shape calculation unit 102 estimates a processing region based on the previous distant traveling road shape. That is, a second estimation processing area is set. In this embodiment, this estimation is realized by executing step S275 of the far road shape calculation process in the first embodiment.

Next, the distant shape calculation unit 102 calculates the likelihood of the overlapping region between the first estimation processing region and the second estimation processing region. The likelihood of the overlapping region is derived based on Bayesian estimation using the likelihood of the first estimation processing region and the likelihood of the second estimation processing region. Specifically, it is calculated based on the following formula (1). In the following formula (1), X represents the likelihood of the overlapping region. A variable A indicates the likelihood of the first estimation processing region. Variable B indicates the likelihood of the second estimation processing region.
X = A * B / {A * B + (1-A) * (1-B)} (1)

  For example, when the evaluation value is E1 shown in FIG. 15, 0.7 is substituted as the variable A in the above formula (1) for the overlapping region of the first estimation processing region and the second estimation processing region, As the variable B, 0.3 is substituted into the above equation (1). Therefore, 0.5 is calculated as the integrated likelihood X for such overlapping regions.

  Next, the distant shape calculation unit 102 sets a region having a likelihood equal to or greater than a predetermined threshold among the regions (pixels) in the imaging apparatus as the current processing region. For example, when the predetermined threshold is set to 0.3 and the evaluation value is E1 shown in FIG. 15, the first estimation processing area (likelihood: 0.7) and the second estimation processing area ( A region obtained by combining the likelihood (0.3) and the overlapping region (likelihood: 0.5) of these two estimation processing regions is set as the current processing region. In addition, about the area | region which does not correspond to any of a 1st estimation process area and a 2nd estimation process area, likelihood is set to 0 and is not set as this process area.

  As shown in FIG. 13, after the execution of such step S410, the above-described steps S220 to S255 are executed.

  The travel path recognition apparatus of the sixth embodiment described above has the same effect as the travel path recognition apparatus 100 of the first embodiment. In addition, since the evaluation value is set higher as the nearby runway distance is longer, and as a result, the likelihood of the neighborhood runway shape is set higher, the processing area estimated based on the neighborhood runway shape is It can be easily set (easily used) as a processing area. Therefore, when it is anticipated that the nearby runway distance is long and the calculation accuracy of the neighborhood runway shape is high, the processing region estimated based on the nearby runway shape can be easily set as the current processing region, and the processing region specifying accuracy ( It is possible to suppress a decrease in the certainty as an area including a far road. In the sixth embodiment, the distant shape calculation unit 102 also corresponds to the evaluation unit in the claims.

G. Seventh embodiment:
The road recognition device 100b of the seventh embodiment is different from the road recognition device 100 of the first embodiment in that it functions as an object detection unit 107 and a stop determination unit 108, as shown in FIG. Since the other device configuration of the runway recognition apparatus 100b of the seventh embodiment is the same as that of the runway recognition apparatus 100 of the first embodiment, detailed description thereof is omitted.

  The object detection unit 107 detects the presence or absence of another vehicle in front of the host vehicle. In the present embodiment, the vehicle position determination unit 106 detects the presence or absence of another vehicle in front of the host vehicle by analyzing the captured image obtained by the imaging unit 21. The stop determination unit 108 determines whether or not the host vehicle is stopped at a red light. In the present embodiment, the stop determination unit 108 determines whether the signal is a red signal by analyzing the captured image obtained by the imaging unit 21 and stops based on the moving speed of the host vehicle obtained from the vehicle speed sensor 22. It is determined whether it is in the middle.

  The road recognition device 100b of the seventh embodiment is different from the road recognition device 100 of the first embodiment in that the execution determination process shown in FIG. 17 is executed. The runway recognition apparatus 100b of the seventh embodiment executes the nearby runway shape calculation process and the target passing point setting process in the same procedure as the runway recognition apparatus 100 of the first embodiment. Moreover, the runway recognition apparatus 100b performs a far road shape calculation process in the same procedure as the runway recognition apparatus 100 of 6th Embodiment.

  The execution determination process is a process for determining whether or not to execute the near road shape calculation process, and is executed when the ignition of the host vehicle is turned on. Therefore, in the seventh embodiment, the neighboring road shape calculation process is not periodically executed. As shown in FIG. 17, the neighborhood shape calculation unit 101 determines whether there is no preceding vehicle in the traveling direction of the host vehicle and whether the host vehicle is stopped by a red signal (step S505). Specifically, the neighborhood shape calculation unit 101 determines that there is no preceding vehicle in the traveling direction of the host vehicle and the host vehicle is red based on the detection result by the object detection unit 107 and the determination result by the stop determination unit 108. It is determined whether or not the vehicle is stopped by a signal.

  When it is determined that there is no preceding vehicle in the traveling direction of the host vehicle and the host vehicle is stopped due to a red signal (step S505: YES), the periodic execution timing of the nearby runway shape calculation process arrives Even so, the processing area specified before the vehicle stops is specified as the current processing area without executing the near road shape calculation process (step S510). In this embodiment, the processing area specified immediately before the stop is specified as the current processing area as the processing area specified before the stop. Instead of the processing area specified immediately before stopping, a processing area specified a predetermined number of times before immediately before stopping may be specified as the current processing area.

  When it is determined that “the preceding vehicle does not exist in the traveling direction of the host vehicle and the host vehicle is stopped due to a red signal” (step S505: NO), the vicinity shape calculation unit 101 and the distant shape calculation unit 102 Executes the above-described near-runway calculation process and the far-off road shape calculation process (step S520).

  In addition, when above-mentioned step S510 is performed, step S220-S255 are performed as a far road shape calculation process.

  The travel path recognition apparatus 100b of the seventh embodiment described above has the same effect as the travel path recognition apparatus 100 of the first embodiment. In addition, when there is no preceding vehicle in the traveling direction of the host vehicle and the host vehicle is stopped due to a red signal, the processing region specified immediately before the stop is specified as the current processing region. , It is possible to suppress a decrease in the accuracy of the processing area. When there is no other vehicle in front of the vehicle and the vehicle is stopped with a red light, the recognition accuracy of the nearby runway shape is the same as that before the vehicle stops. On the other hand, when there is a preceding vehicle and the vehicle is stopped at a red light, for example, the case where the paint of the vehicle body of the preceding vehicle is erroneously recognized as a lane marking (neighboring runway shape) is assumed, and the characteristic accuracy of the processing region decreases. There is a fear. Therefore, according to the runway recognition apparatus 100b of the present embodiment, it is possible to suppress a decrease in the processing area identification accuracy.

H. Eighth embodiment:
As shown in FIG. 18, the travel path recognition apparatus 100 c according to the eighth embodiment is electrically connected to the millimeter wave radar 24 mounted on the host vehicle, and functions as the three-dimensional structure detection unit 109. However, it differs from the runway recognition apparatus 100 of 1st Embodiment. Since the other configuration of the road recognition device 100c according to the eighth embodiment is the same as the device configuration of the road recognition device 100 according to the first embodiment, the same components are denoted by the same reference numerals, and detailed description thereof is provided. Is omitted.

  The millimeter wave radar 24 detects an object in front of the host vehicle using millimeter wave radio waves. More precisely, the object is detected as a plurality of detection points (targets). The three-dimensional structure detection unit 109 is based on the detection result by the millimeter wave radar 24 and the captured image obtained by the imaging unit 21, and the three-dimensional structure (preceding vehicle, oncoming vehicle, vehicle on the intersecting road, median strip, Pedestrians, guardrails, manholes, traffic lights, etc.), the distance between the three-dimensional structure and the vehicle, the position of the three-dimensional structure, the size of the three-dimensional structure, and the relative speed of the three-dimensional structure with respect to the own vehicle. To detect.

  The runway recognition apparatus 100c according to the eighth embodiment executes a nearby runway shape calculation process, a far road form calculation process, and a target passing point setting process in the same procedure as the runway recognition apparatus 100 according to the first embodiment. However, the specific procedure of step S215 of the far road shape calculation process shown in FIG. 4 is slightly different from that of the first embodiment.

  In step S215 in the first embodiment, the distance between the adjacent tracks (division line) is extended to estimate the far track shape, and the processing region is set based on the far track shape. In contrast, in the eighth embodiment, the distant shape calculation unit 102 exists on the side edge of the road detected by the three-dimensional structure detection unit 109 in the processing region set in the same manner as in the first embodiment. A processing area is set inside an area partitioned by a three-dimensional structure (hereinafter referred to as “roadside object”).

  In the example of FIG. 19, the host vehicle 500 travels from the left side to the right side in FIG. 19 on the travel lane Ln3, and is positioned in front of the intersection CR2. At this time, in the travel lane Ln3, the two regions Er3 and El3 beyond the intersection CR2 are temporarily specified as the processing regions based on the far travel road shape. Of these two areas Er3 and El3, the area in the area LW is finally set as a processing area. The processing area LW includes a roadside object PgR which is a three-dimensional structure detected along the travel lane Ln3 in the vicinity of the travel lane Ln3, and a three-dimensional structure detected along the travel lane Ln3 in the vicinity of the adjacent travel lane Ln4. It is specified as an area partitioned by the roadside object PgL. The roadside object PgR corresponds to, for example, a so-called central separation zone located at a boundary with an opposite lane (not shown). Further, the roadside object PgL corresponds to a guardrail located at the boundary between the travel lane Ln4 and a sidewalk (not shown).

  The runway recognition apparatus of the eighth embodiment described above has the same effect as the runway recognition apparatus 100 of the first embodiment. In addition, since the processing region is set inside the region partitioned by the inside of the roadside object detected by the three-dimensional structure detection unit, it is possible to suppress a decrease in the accuracy of specifying the processing region. This is because, in general, the runway exists inside an area defined by roadside objects.

I. Ninth embodiment:
As illustrated in FIG. 20, the travel path recognition device 100 d of the ninth embodiment is electrically connected to the connection interface unit (connection IF unit) 25, as the travel lane identification unit 110 and the travel lane estimation unit 111. It differs from the runway recognition device 100c of the eighth embodiment in that it functions. Since the other configuration of the road recognition device 100d of the ninth embodiment is the same as the device configuration of the road recognition device 100c of the eighth embodiment, the same components are denoted by the same reference numerals, and detailed description thereof will be given. Is omitted.

  The connection IF unit 25 is electrically connected to the navigation device 600 mounted on the host vehicle, and acquires information indicating the position of the host vehicle and map information from the navigation device 600. Such map information includes information indicating the driving lane of each road. The travel lane identification unit 110 identifies the current travel lane of the host vehicle based on information about the current travel lane input from the navigation device 600 via the connection IF unit 25. The travel lane estimation unit 111 estimates a travel lane after passing, which is a lane in which the host vehicle travels after passing through the intersection when the host vehicle reaches the intersection. Specifically, the traveling lane estimation unit 111 identifies a roadside object existing ahead of the current traveling lane among the three-dimensional structures detected by the three-dimensional structure detection unit 109, and identifies the identified roadside object and the current roadside object. Based on the travel lane, the post-passage travel lane is estimated from the travel lanes existing ahead of the current position of the host vehicle obtained from the map information (the destination that has passed the intersection). For example, in the example of FIG. 19, the own vehicle 500 is currently traveling in the travel lane Ln3, but there is an intersection CR2, and therefore, the travel lane scheduled to travel in the future cannot be specified in the analysis of the captured image. However, each information that the own vehicle 500 is currently traveling in the travel lane Ln3, that there is a roadside object PgR ahead, and that there is a travel lane Ln3 and a travel lane Ln4 ahead of the intersection CR2. Therefore, after passing the intersection CR2, the host vehicle 500 is predicted to travel on the travel lane Ln3, and is estimated as the travel lane after passing the travel lane Ln3.

  In the same manner as in the eighth embodiment, the distant shape calculation unit 102 detects the roadside detected by the three-dimensional structure detection unit 109 among the processing regions (temporary processing regions) set in the same manner as in the first embodiment. Further, an area within the travel lane that is inside the area partitioned by the object is set as a final processing area. Therefore, in the ninth embodiment, unlike the eighth embodiment, an area outside the travel lane Ln3 in the area El3 shown in FIG. 19 is not set as a processing area.

  The travel path recognition device 100d of the ninth embodiment described above has the same effect as the travel path recognition device 100c of the eighth embodiment. Furthermore, since the post-passage travel lane is estimated and the processing region is further limited by using the post-passage travel lane, it is possible to further suppress a reduction in the accuracy of the processing region. In particular, even if there is an intersection and the calculation accuracy of the nearby runway shape is low, it is possible to suppress a decrease in the processing region specifying accuracy.

J. et al. Tenth embodiment:
As illustrated in FIG. 21, the travel path recognition apparatus 100 e according to the tenth embodiment includes a point electrically connected to the acceleration sensor 26, a point that functions as the posture change detection unit 112, and a correction amount map storage unit 120. In the point provided, it differs from the runway recognition apparatus 100 of 1st Embodiment. Since the other configuration of the road recognition device 100e of the tenth embodiment is the same as the device configuration of the road recognition device 100 of the first embodiment, the same components are denoted by the same reference numerals, and detailed description thereof will be given. Is omitted.

  The acceleration sensor 26 detects the acceleration of the host vehicle. The posture change detection unit 112 detects a change in the posture of the host vehicle. In the present embodiment, the posture change detection unit 112 detects a change in the posture of the host vehicle based on the acceleration obtained by the acceleration sensor 26. Specifically, when the acceleration is a positive value in the traveling direction, that is, when the host vehicle is accelerating, the posture of the host vehicle is such that the front side is directed vertically upward and the rear side is directed vertically downward. It detects when it changes. Such a change in posture is called a so-called nose-up. Further, when the host vehicle is decelerating, the posture of the host vehicle is detected when the front side changes so as to face the vertically lower side and the rear side faces the vertically upper side. Such a change in posture is called a so-called nose down.

  The correction amount map storage unit 120 stores a correction amount map in advance. The correction amount map is a map in which the moving speed (vehicle speed) of the host vehicle, the acceleration of the host vehicle, and the correction amount are associated with each other. The correction amount is a correction amount for correcting the position of the processing area set in step S215 of the far road calculation process, and is stored together with the correction direction in units of the number of pixels (pixels). When the above-described nose-up occurs, the imaging direction of the imaging unit 21 changes, so that the imaging range that appears in the captured image is shifted upward from the imaging range that was set during the initial calibration. For this reason, the processing region set in step S215 is detected by being shifted upward with respect to the actual region. On the other hand, when a nose-down occurs, the imaging range that appears in the captured image is shifted downward from the imaging range that was set during the initial calibration. For this reason, the processing region set in step S215 is detected by being shifted downward with respect to the actual region. In view of this, in step S215, the distant shape calculation unit 102 obtains a processing region according to the above-described procedure, and corrects the processing region so as to cancel the shift amount due to the change in the posture of the host vehicle. Since the correction amount at this time correlates with the vehicle speed and the acceleration, a correction amount map is created by obtaining the vehicle speed, acceleration, and correction amount in advance through experiments, and stored in the correction amount map storage unit 120. Accordingly, the distant shape calculation unit 102 specifies the correction amount with reference to the correction amount map based on the vehicle speed obtained from the vehicle speed sensor 22 and the acceleration obtained from the acceleration sensor 26.

  The travel path recognition apparatus 100e of the tenth embodiment described above has the same effect as the travel path recognition apparatus 100 of the first embodiment. In addition, when the posture change of the host vehicle is detected, the position of the processing region is corrected and specified according to the posture change, so that the region that is not appropriate with the posture change, for example, a region that does not include a distant road Can be specified as a processing region. In addition, since the correction amount for correcting the processing area is set with reference to a correction amount map stored in advance, the correction amount can be set easily and in a short time.

K. Eleventh embodiment:
As illustrated in FIG. 22, the road recognition device 100 f according to the eleventh embodiment is configured to recognize the road according to the tenth embodiment in that it functions as the template image specifying unit 113 and does not include the correction amount map storage unit 120. It is different from the device configuration of the device 100e. Since the other components in the road recognition device 100f of the eleventh embodiment are the same as those of the road recognition device 100e of the tenth embodiment, the same components are denoted by the same reference numerals and detailed description thereof is omitted. To do. The template image specifying unit 113 specifies a template image used when setting the processing area. Such a template image will be described later.

  The track recognition device 100f of the eleventh embodiment is different from the track recognition device 100e of the tenth embodiment in the procedure of the far track shape calculation process. The runway recognition apparatus 100f according to the eleventh embodiment performs a nearby runway shape calculation process and a target passing point setting process in the same procedure as the runway recognition apparatus 100e according to the tenth embodiment.

  As shown in FIG. 23, when it is determined in step S205 that there is an input of a nearby runway shape (step S205: YES), the posture change detection unit 112 changes the posture of the host vehicle, specifically, nose-up. Alternatively, it is determined whether or not nose down has occurred (step S440). When it is determined that nose-up or nose-down has occurred (step S440: YES), the template image specifying unit 113 sets a template image including a processing region set before nose-up or nose-down occurs ( Step S445). Specifically, in the captured image used when the previous processing area is set (that is, the captured image obtained last time), the area including the processing area, for example, the processing area Er1 and the processing area El1 in FIG. An image of a rectangular area including both is cut out and set as a template image.

  The far shape calculation unit 102 executes pattern matching of the template image set in step S445 in the current captured image (step S450). That is, an area having the same size as the template image set in step S445 is specified while being shifted pixel by pixel in the captured image obtained this time, and the similarity between the area and the template image is obtained.

  The distant shape calculation unit 102 sets a processing region in the most similar image region as a result of step S450 (step S455). Specifically, the distant shape calculation unit 102 sets an area having the same position as the relative position of the processing area in the template image as the processing area in the most similar image area. After execution of step S445, the above-described steps S220 to S255 are executed.

  When it is determined in step S440 described above that no nose up or nose down occurs (step S440: NO), step S210 described above is executed.

  The track recognition device 100f of the eleventh embodiment described above has the same effect as the track recognition device 100e of the tenth embodiment. In addition, when a nose-up or nose-down occurs, an image of the region including the processing region in the captured image obtained before the occurrence is set as a template image, and pattern matching is performed on the captured image obtained this time with the template image. Since the processing area is set by performing the above, it is not necessary to correct the processing area when nose up or nose down occurs. Further, since it is not necessary to store the correction amount at the time of correction in advance, the storage capacity required for such storage can be omitted.

L. Twelfth embodiment:
The runway recognition apparatus 100g of the twelfth embodiment differs from the runway recognition apparatus 100c of the eighth embodiment shown in FIG. 18 in that it functions as a correction amount specifying unit 114 as shown in FIG. Since the other components in the road recognition device 100g of the twelfth embodiment are the same as those of the road recognition device 100c of the eighth embodiment, the same components are denoted by the same reference numerals and detailed description thereof is omitted. To do. The correction amount specifying unit 114 specifies a correction amount used when correcting the position of the processing region. In the runway recognition apparatus 100g, the position of the processing region is corrected in accordance with the change in the posture of the host vehicle, similarly to the runway recognition apparatus 100e in the tenth embodiment. In the road recognition device 100e of the tenth embodiment, the correction amount is specified by referring to the correction amount map based on the vehicle speed and acceleration. However, the road recognition device 100g of the twelfth embodiment is obtained by the imaging unit 21. The correction amount is specified using the captured image and the measurement result of the millimeter wave radar 24.

  The travel path recognition apparatus 100g of the twelfth embodiment differs from the travel path recognition apparatus 100 of the first embodiment in the procedure of the far travel road shape calculation process. Note that the runway recognition apparatus 100g of the twelfth embodiment performs a nearby runway shape calculation process and a target passing point setting process in the same procedure as the runway recognition apparatus 100 of the first embodiment.

  As shown in FIG. 25, when it is determined in step S205 that there is an input of a nearby runway shape (step S205: YES), the correction amount specifying unit 114 receives from the host vehicle up to a specific point in the previous captured image. The difference between the distance and the distance from the host vehicle to the specific point in the current captured image is calculated (step S460). The specific position can be set to an arbitrary position as long as it is a position in an actual environment that is the same between the previous time and the current time. For example, the position of a pole (for example, a pole for posting a sign) arranged on the roadside may be set as the specific position.

  Next, the correction amount specifying unit 114 calculates the distance to the specific three-dimensional structure based on the previous measurement result of the millimeter wave radar 24 and the distance to the specific three-dimensional structure based on the measurement result of the current millimeter wave radar 24. The difference is calculated (step S465). The specific three-dimensional structure can be defined as an arbitrary three-dimensional structure as long as it is a three-dimensional structure in an actual environment that is the same between the previous time and the current time. Specific examples of steps S460 and S465 described above will be described with reference to FIGS. 26, 27, and 28. FIG.

  In the example shown in FIG. 26, the posture of the host vehicle 500 is a normal posture, that is, a posture in which the longitudinal direction of the vehicle is parallel to the horizontal direction. At this time, the distance from the own vehicle 500 to the position P2 of the pole PL existing in front of the own vehicle 500 is estimated as the distance L1 based on the captured image. In this embodiment, this distance is obtained from the relative position P2 of the pole PL in the captured image. Since the imaging range Ag is a predetermined range at the normal posture, that is, the posture at the time of executing the calibration of the imaging unit 21, it is determined from the own vehicle 500 depending on the relative position of the position P <b> 2 in the captured image. The distance to the position P2 can be obtained (estimated). In the example of FIG. 26, the distance to the pole PL is estimated as the distance L2 based on the measurement result of the millimeter wave radar 24. Since the millimeter wave radar 24 irradiates the millimeter wave forward and receives the reflected wave from the pole PL, the millimeter wave radar 24 strikes the pole PL in the horizontal direction from the position of the millimeter wave radar 24 as the distance from the own vehicle 500 to the pole PL. A distance L2 to the position P1 is obtained.

  As shown in FIG. 27, for example, when the host vehicle 500 suddenly brakes after the state of FIG. 26, nose-down occurs. In this case, the imaging range Ag is directed vertically downward compared to the case of FIG. For this reason, the relative position of the pole PL in the captured image changes vertically upward, and appears as if the pole PL is located farther away, as indicated by a broken line in FIG. Therefore, in the example of FIG. 27, the position of the pole PL is specified as a position P2a farther from the host vehicle 500, unlike the position P2 shown in FIG. At this time, the distance between the host vehicle 500 and the position P2a is estimated as the distance L11. Then, the difference between the distance L11 shown in FIG. 27 and the distance L1 shown in FIG. 26 is obtained in the above-described step S460. This difference includes the travel distance of the host vehicle 500 until the state of FIG. 26 changes to the state of FIG. 27 and the relative displacement of the pole PL in the captured image caused by the nose-down described above. It is.

  On the other hand, as shown in FIG. 28, after the state of FIG. 26, when the own vehicle 500 suddenly brakes and nose-down occurs as in FIG. 27, according to the measurement result of the millimeter wave radar 24, the pole PL Accordingly, the distance L12 from the own vehicle 500 to the pole PL measured at this time is also obtained as the distance from the own vehicle 500 to the position P1 of the pole PL as in FIG. Then, the difference between the distance L12 shown in FIG. 28 and the distance L2 shown in FIG. 26 is obtained in step S465. This difference is substantially equal to the travel distance of the host vehicle 500 until the state of FIG. 26 changes to the state of FIG.

  As shown in FIG. 25, the correction amount specifying unit 114 specifies the correction amount so as to cancel the difference between the two types of difference, that is, the difference calculated in step S460 and the difference calculated in step S465 (step S470). Specifically, the correction amount is specified so that the difference obtained in step S465 is subtracted from the difference obtained in step S460, and the remainder is canceled. For example, in the example of FIGS. 26 to 28 described above, the difference between the distance L12 and the distance L2 is subtracted from the difference between the distance L11 and the distance L1. If the difference is positive, that is, if the value is such that the pole PL is located further away, it is located closer, that is, positioned lower in the captured image. The correction amount is specified so that

  As shown in FIG. 25, the far-distance shape calculation unit 102 estimates the far-run road shape by extending the near-run road (parting line) using the input near-run road shape (step S475). This step S475 is the same as step S210 described above. The far shape calculation unit 102 sets a temporary processing region based on the estimated far road shape (step S480). The provisional processing area means a processing area before correction by the correction amount specified in step S470. Step S480 is the same as step S215 described above. The distant shape calculation unit 102 sets the processing region by shifting the temporary processing region set in step S480 by the correction amount specified in step S470 (step S485). After execution of step S485, the above-described steps S220 to S255 are executed.

  The road recognition device 100g of the twelfth embodiment described above has the same effect as the road recognition device 100g of the tenth embodiment. In addition, the correction amount for correcting the position of the processing region when nose-up or nose-down occurs is calculated using the captured image obtained by the imaging unit 21 and the measurement result of the millimeter wave radar 24. The correction amount can be obtained with high accuracy.

M.M. Thirteenth embodiment:
The runway recognition apparatus 100h according to the thirteenth embodiment differs from the runway recognition apparatus 100f according to the eleventh embodiment in that it functions as a template image correction unit 115, as shown in FIG. Since the other components in the road recognition device 100h according to the thirteenth embodiment are the same as those of the road recognition device 100f according to the eleventh embodiment, the same components are denoted by the same reference numerals, and detailed description thereof is omitted. To do. The runway recognition device 100h executes a nearby runway shape calculation process, a far road shape calculation process, and a target passing point setting process in the same procedure as the runway recognition apparatus 100f of the eleventh embodiment.

  The template image correction unit 115 corrects the template image in step S445 of the far road shape calculation process shown in FIG. Specifically, based on the mounting parameter correlated with the mounting position of the imaging unit 21 or the attitude of the imaging unit 21, the parameter correlated with the shape of the traveling lane of the host vehicle, and the parameter correlated with the behavior of the host vehicle, Correct the image. For example, the mounting parameters include the yaw, pitch, and roll in the imaging direction of the imaging unit 21 and the relative position (coordinates) of the imaging unit 21 from the center of gravity of the host vehicle. As parameters correlating with the shape of the traveling lane, each parameter indicating the shape of a nearby traveling road corresponds. The parameters correlated with the behavior of the host vehicle include, for example, the presence or absence of a lane change, the schedule of a right or left turn, the vehicle speed and acceleration / deceleration of the host vehicle.

  The travel path recognition device 100h of the thirteenth embodiment described above has the same effect as the travel path recognition device 100f of the eleventh embodiment. In addition, the template image is corrected based on the mounting parameters correlated with the mounting position or orientation of the imaging device, the parameters correlated with the shape of the vehicle's travel lane, and the parameters correlated with the behavior of the vehicle. Even if each parameter changes, an appropriate image can be used as a template image.

N. Fourteenth embodiment:
Since the device configuration of the track recognition device of the fourteenth embodiment is the same as the device configuration of the track recognition device 100f of the eleventh embodiment, the same components are denoted by the same reference numerals, and detailed description thereof is omitted. To do.

  The runway recognition apparatus 100f according to the fourteenth embodiment performs a nearby runway shape calculation process, a far roadform shape calculation process, and a target passing point setting process in the same procedure as the runway recognition apparatus 100f according to the eleventh embodiment. However, the far road shape calculation process of the fourteenth embodiment is different from the far road shape calculation process of the eleventh embodiment in step S450 in the far road shape calculation process shown in FIG.

  As shown in FIG. 30, in the pattern matching process in the fourteenth embodiment, the distant shape calculation unit 102 is a region to be compared with the template image in the current captured image (hereinafter referred to as “comparison target region”). Is specified (step S705). The distant shape calculation unit 102 specifies a broken line-containing region in the template image set in step S445 shown in FIG. 23 (step S710). The broken line-containing region is a region including a broken line. For example, when the dividing line is a broken line, it means a region including the dividing line. As will be described later, the distant shape calculation unit 102 compares each pixel in the comparison target region with each pixel at the same position in the template image and gives a score according to whether or not they match. At this time, it is determined whether or not a pixel to be compared in the comparison target region (hereinafter referred to as “comparison pixel”) corresponds to a pixel in the broken line-containing region in the template image (step S715).

  When it is determined that the comparison pixel is a pixel corresponding to a pixel in the broken line-containing region (step S715: YES), the distant shape calculation unit 102 determines that the comparison pixel and the corresponding pixel in the template image (pixels having the same relative position) Are compared to determine whether or not they match (step S720). When it is determined that they match (step S720: YES), the distant shape calculation unit 102 gives a score “1” to the comparison pixel (step S725). On the other hand, when it is determined that they do not match (step S735: NO), a score “0.5” is given to the comparison pixel (step S730).

  In the above-described step S715, when it is determined that the comparison pixel is not a pixel corresponding to the pixel in the broken line-containing region (step S715: NO), the comparison pixel and the corresponding pixel (pixels having the same relative position) in the template image are determined. It is determined by comparison whether or not they match (step S735). Such step S735 is the same as step S720 described above. When it is determined that they match (step S735: YES), the distant shape calculation unit 102 gives a score “1” to the comparison pixel as in step S725 described above (step S740). On the other hand, when it is determined that they do not match (step S735: NO), unlike the above-described step S730, a score “0” is given to the comparison pixel (step S745).

  After any of the above-described steps S725, S730, S740, and S745 is executed, the distant shape calculation unit 102 determines whether or not the comparison has been completed for all the pixels in all the comparison target regions (step S750). If it is determined that the comparison has not ended (step S750: NO), the process returns to step S705 described above. On the other hand, when it is determined that the comparison has been completed (step S750: YES), the far shape calculation unit 102 calculates and stores the total score for each comparison target region (step S755). The total score is a value obtained by adding the scores given by executing any one of steps S725, S730, S740, and S745 for each pixel in each comparison target region. The higher the total score, the more similar to the template image. Then, after execution of step S755, the pattern matching process ends, and step S455 shown in FIG. 23 is executed. At this time, the region with the highest total score is the most similar region, and the processing region is set in this region.

  Here, as described above, when the comparison pixel is a pixel corresponding to the pixel in the broken line-containing region, the score given when the comparison pixel does not match the corresponding pixel is “0.5”. The score is higher than the score “0” when the comparison pixel is not a pixel corresponding to the pixel in the broken line-containing region. By doing in this way, when the lane marking of the traveling lane of the own vehicle is a broken line, the influence of the result of the pattern matching process on the total score can be reduced compared to the case where the lane marking is a solid line. In general, when the lane marking is a broken line, even if it is the same lane marking, there is a high possibility that the mode shown in the previous captured image is different from the mode shown in the current captured image. Specifically, since the position of the boundary between the painted portion and the unpainted portion of the broken line is shifted as the host vehicle travels, even if it is the same marking line, it is reflected in the previous captured image. There is a high possibility that the mode and the mode shown in the current captured image are different. Therefore, in the case of the broken line, the same partition line can be evaluated as the same by reducing the influence of the pattern matching result on the total score.

  100f of 14th Embodiment demonstrated above has an effect similar to the track recognition apparatus 100f of 11th Embodiment. In addition, when the lane marking is a broken line, the effect of the pattern matching result on the score indicating the degree of coincidence is reduced compared to the case where the lane marking is a solid line, thereby suppressing a decrease in the accuracy of processing area specification. it can. When the lane marking is a broken line, the shape is likely to be different at different imaging timings even though it is the same line. Therefore, in this case, since the influence on the score indicating the degree of coincidence (total score) is reduced, the possibility of recognizing the same lane line as the same lane line can be increased.

O. Fifteenth embodiment:
The road recognition device 100i of the fifteenth embodiment differs from the road recognition device 100 of the first embodiment in that it functions as an intersection determination unit 116 and a direction identification unit 117, as shown in FIG. Since the other structure of the road recognition apparatus 100i of 15th Embodiment is the same as the road recognition apparatus 100 of 1st Embodiment, the detailed description is abbreviate | omitted.

  The intersection determination unit 116 determines whether or not the host vehicle is traveling in the intersection area or stopped in the intersection area. The direction specifying unit 117 detects a plurality of paints that form a stripe pattern (so-called zebra) on a pedestrian crossing on the traveling lane of the host vehicle, and specifies the longitudinal directions of the detected plurality of paints. In general, the longitudinal direction of the plurality of paints is a direction along the traveling lane.

  As shown in FIG. 32, the far road shape calculation process of the fifteenth embodiment is the same as the far road shape calculation process of the first embodiment shown in FIG. 4 in that steps S280 to S284 are executed instead of step S275. Different, the other procedures are the same. Note that the neighborhood road shape calculation process and the target passing point setting process of the fifteenth embodiment are the same as those of the first embodiment.

  In step S270 described above, when it is determined that there is a history of the previous far road shape (step S270: YES), the intersection determination unit 116 is traveling in the intersection area or is stopping in the intersection area. It is determined whether or not there is (step S280). The definition of the intersection area is as described in the fourth embodiment. In addition, the determination as to whether or not the host vehicle is traveling or stopped in the intersection can be realized based on the vehicle speed measured by the vehicle speed sensor 22 in addition to the same procedure as step S206 of the fourth embodiment. When it is determined that the host vehicle is not traveling in the intersection area or is not stopped in the intersection area (step S280: NO), the far road shape calculation process ends.

  When it is determined that the host vehicle is traveling in the intersection area or stopped in the intersection area (step S280: YES), the far-field shape calculation unit 102 crosses the traveling lane of the host vehicle in the current captured image. It is determined whether or not there are a plurality of paints constituting the sidewalk stripe pattern (step S281). When it is determined that there is no plurality of paints (step S281: NO), the far road shape calculation process ends. On the other hand, when it is determined that there are a plurality of paints (step S281: YES), the direction specifying unit 117 of the vehicle of the edges along the longitudinal direction of the plurality of paints specified in step S281. An edge located at the right end and an edge located at the left end in the travel lane are specified (step S282). Based on the two edges specified in step S282, the direction specifying unit 117 specifies the position of a point (so-called vanishing point) where virtual lines intersect when extending the edges in the current captured image (step). S283). The distant shape calculation unit 102 sets a processing region based on the position of the vanishing point specified in step S283 (step S284).

  In the example of FIG. 33, it is determined that there are a plurality of paints forming the stripe pattern of the pedestrian crossing Pd3 in the travel lane Ln5 of the host vehicle. The travel lane Ln5 is partitioned by the lane markings Lr4 and Ll4 that are slightly visible below the captured image F20. In step S282, among the plurality of edges, the inner edge Eg1 of the right paint Zr partially included in the travel lane Ln5, and the left paint partially included in the travel lane Ln5. An inner edge Eg2 of Zl is specified. In step S283, a point P10 where two virtual lines Vl1 and Vl2 obtained by extending the two edges Eg1 and Eg2 identified in step S282 intersect is identified as a vanishing point.

  Using the example of FIG. 33, the method for setting the processing area in step S284 will be specifically described. First, two first points on the two virtual lines Vl1 and Vl2, which are two first points before the first distance L21 determined in advance from the point P10 (intersection position of the two virtual lines Vl1 and Vl2). Points P1r and P1l are specified. Next, two points that are different from the two first points P1r and P1l on the two virtual lines Vl1 and Vl2, and that are farther from the end on the far side of the pedestrian crossing Pd3 by a predetermined second distance L22. The side points are specified as the second points P2r and P2l. Then, along the virtual line Vl1, a rectangle having a predetermined third distance L3 as a width in the direction intersecting the travel lane Ln5, with the line segment connecting the first point P1r and the second point P2r as the center. Is set as the processing area Er4. Similarly, a rectangular area having a width in the direction intersecting the travel lane Ln5 with the third distance L3 centered on a line segment connecting the first point P1l and the second point P2l along the virtual line Vl2. And set as the processing area El4.

  As shown in FIG. 32, after the execution of step S284, the above-described steps S220 to S255 are executed.

  The travel path recognition apparatus 100i of the fifteenth embodiment described above has the same effect as the travel path recognition apparatus 100 of the first embodiment. In addition, when it is determined that the host vehicle is traveling or stopped in the intersection area, the processing area is set using the longitudinal direction of the paint that forms the stripe pattern of the pedestrian crossing. Decrease in specific accuracy can be suppressed. In general, since the longitudinal direction of a plurality of paints forming the striped pattern of the pedestrian crossing on the traveling lane is parallel to the traveling lane, the processing region is set by using the longitudinal direction, thereby specifying the processing region. It is because the fall of can be suppressed.

P. Sixteenth embodiment:
The runway recognition apparatus 100j of the sixteenth embodiment differs from the apparatus configuration of the runway recognition apparatus 100 of the first embodiment in that it functions as a normality determination unit 118, as shown in FIG. Since the other components in the runway recognition apparatus 100j of the sixteenth embodiment are the same as those of the runway recognition apparatus 100 of the first embodiment, the same components are denoted by the same reference numerals and detailed description thereof is omitted. To do. The normality determination unit 118 determines the normality of the white line candidate temporarily selected in the far road shape calculation process in the sixteenth embodiment described later.

  As shown in FIG. 35, the far road shape calculation process of the sixteenth embodiment executes steps S805 to S810 instead of steps S225 to S240 as compared to the far road calculation process of the first embodiment shown in FIG. It differs only in respect. In addition, the runway recognition apparatus 100j of 16th Embodiment performs a near runway shape calculation process and a target passing point setting process in the procedure similar to the runway recognition apparatus 100 of 1st Embodiment. In the first embodiment, the Hough transform is performed in order to identify the white line candidate in the processing area. However, in the sixteenth embodiment, LMedS (Least Median Squares) is used. LMedS is a kind of so-called random sampling method. Note that RANSAC (RANdom SAmple Consensus) may be used instead of LMedS.

  As shown in FIG. 35, after execution of step S220, the distant shape calculation unit 102 determines a white line candidate using LMedS (step S805). Specifically, any two of the edges in the processing region extracted in step S220 are sampled randomly. Next, the square of the error between all the other edges (for example, N) and the straight line is calculated from the straight line equation connecting the two edges. Next, the obtained median value of the N square errors is specified and recorded as the evaluation score of the straight line. The above is executed for two combinations of all edges. And the straight line of the lowest evaluation score is determined as a white line candidate among all the obtained evaluation scores. The white line candidate is calculated as a far road shape when it is determined to be normal as will be described later, and can therefore be referred to as a “provisional far road shape”.

  The normality determination unit 118 determines whether the white line candidate determined in step S805 is normal (step S810). A specific method for determining the normality will be described with reference to FIG. In FIG. 36, for the four cases C1 to C4, white line candidates m1 to m4, error distribution, and evaluation scores S1 to S4 for each white line candidate are shown.

  In case C1, the evaluation score S1 is a relatively small value, and is determined to be normal in the present embodiment. On the other hand, in the other three cases C2 to C4, it is determined that it is not normal, that is, abnormal. Specifically, in case C2, since the evaluation score is relatively large, it is determined to be abnormal. In case C3, the evaluation score S3 is relatively small. However, in case C3, there is a continuous error range in which an error (square error) of a value deviating from the median value is continuous, and there is a relatively wide error range having a large frequency and equal to each other. Such a situation is assumed as shown in case C3 in which a similar straight line m30 may exist in addition to the straight line m3 determined as the white line candidate. Specifically, when there is a guardrail installed along the white line in the vicinity of the white line (partition line), the straight line formed by the edge of the guardrail also deviates from the median in the error distribution. The error appears in a uniformly distributed manner. In case C4, although the evaluation score S4 is relatively small, the number of errors calculated is small because the total number of edges is small. As a result, it is assumed that the probability that the white line candidate m4 is a white line is low.

In this embodiment, in order to recognize the above three cases C2 to C4 as abnormal, it is not normal (abnormal) when any of the following conditions (i) to (iii) is satisfied. It is determined, and if neither condition is satisfied, it is determined that it is normal.
(I) The median evaluation score in the error distribution obtained by LMedS is larger than a predetermined threshold evaluation score.
(Ii) In an error distribution obtained by LMedS, an error range in which an error of a value deviating from the median value continues, and an error range in which the frequencies are greater than a predetermined threshold frequency and equal to each other is a predetermined range Wider than.
(Iii) The obtained number of errors is smaller than a predetermined threshold number.

  As shown in FIG. 35, when it is determined that the white line candidate is normal as a result of step S810 described above (step S810: YES), the distant shape calculation unit 102 executes steps S245 to S255 described above. On the other hand, when it is determined that the white line candidate is not normal (step S810: NO), the far road shape calculation process ends. Therefore, in this case, the far road shape is not calculated.

  The runway recognition apparatus 100j of the sixteenth embodiment described above has the same effect as the runway recognition apparatus 100 of the first embodiment. In addition, since the far road shape is not calculated when it is determined that the white line candidate determined by LMedS is not normal, it is possible to prevent the far road shape that is not normal from being stored in the far shape storage unit 104 as a history. For this reason, the fall of the precision at the time of setting a process area | region using a far road shape can be suppressed.

  Further, since the normality of the white line candidate is determined based on the error distribution, the normality can be accurately determined. Further, when any of the above conditions (i) to (iii) is satisfied, it is determined that the white line candidate is not normal, and thus the normality of the white line candidate can be determined with high accuracy.

Q. Seventeenth embodiment:
The runway recognition apparatus 100k according to the seventeenth embodiment differs from the apparatus configuration of the runway recognition apparatus 100j according to the sixteenth embodiment in that it functions as a runway estimation unit 119, as shown in FIG. The other components in the track recognition device 100k of the seventeenth embodiment are the same as those of the track recognition device 100j of the sixteenth embodiment, so the same components are denoted by the same reference numerals and detailed description thereof is omitted. To do. The runway estimation unit 119 estimates the current far road shape based on the previous far road shape.

  As shown in FIG. 38, the far road shape calculation process of the seventeenth embodiment differs from the far road calculation process of the sixteenth embodiment shown in FIG. 35 only in that step S800 is additionally executed. In addition, the runway recognition apparatus 100k of 17th Embodiment performs a near runway shape calculation process and a target passing point setting process in the same procedure as the runway recognition apparatus 100 of 1st Embodiment.

  As shown in FIG. 38, after the execution of step S220, the runway estimation unit 119 is based on the previous far runway shape, imaging parameters, and information indicating the movement of the vehicle stored in the far shape storage unit 104. The current far road shape is estimated (step S800). Since this step S800 is the same process as the process of estimating the current far road shape in step S275 described above, detailed description thereof will be omitted. After step S800, steps S805 and S810 described above are executed. However, the specific procedure of step S810 in the present embodiment is different from the procedure of the sixteenth embodiment.

  Specifically, the normality determination unit 118 compares the current far road shape estimated in step S800 with the white line candidate determined in step S805, and the position and angle of these two shapes (straight lines). Find the difference. Only the difference between one of the position and the angle may be obtained. When the difference is equal to or greater than a predetermined threshold, it is determined that the difference is not normal (abnormal), and when the difference is smaller than the threshold, it is determined that the difference is normal.

  In FIG. 39, straight lines Lr4 and Ll4 indicate the neighboring road shape. The straight lines Lr5 and L15 indicate the current far road shape estimated from the previous far road shape. Straight lines Lr6 and Ll6 indicate white line candidate straight lines. As shown in FIG. 39, the straight line Lr5 and the straight line Lr6 intersect each other, and the angle difference is relatively large. Similarly, the straight line L15 and the straight line L16 cross each other, and the angle difference is relatively large. Accordingly, in this case, the white line candidate straight lines Lr6 and L16 are likely to be determined to be abnormal. In addition, the reason why the white line candidate is greatly different from the current far road shape estimated from the previous far road shape in this way is, for example, the white line from the edge extracted from the guardrail or planting existing along the lane line. The case where a candidate is determined is assumed.

  The track recognition device 100k of the seventeenth embodiment described above has the same effect as the track recognition device 100j of the sixteenth embodiment. In addition, when the white line candidate determined by LMedS is compared with the current far road shape estimated from the previous far road shape, the difference between the angle and the position of the two straight lines is equal to or greater than the threshold value. Therefore, it is possible to accurately determine the normality of the white line candidate.

R. Eighteenth embodiment:
Since the device configuration of the track recognition device in the eighteenth embodiment is the same as the device configuration of the track recognition device 100k in the seventeenth embodiment shown in FIG. 37, the same components are assigned the same reference numerals, Detailed description thereof is omitted.

  The far road shape calculation process in the eighteenth embodiment is the same as the far road shape calculation process in the seventeenth embodiment shown in FIG. However, the specific processing in step S805 differs from that in the seventeenth embodiment.

  The traveling path recognition device 100k according to the eighteenth embodiment executes step S805 according to the procedure shown in FIG. Specifically, the normality determination unit 118 selects two edges and specifies a straight line passing through the two edges (step S905). The normality determination unit 118 obtains a temporary error distribution using LMedS for the straight line specified in step S905 (step S910). Note that the term “provisional error distribution” is because such error distribution may be corrected as will be described later. The normality determination unit 118 obtains the degree of coincidence between the current far road shape estimated in step S800 and the straight line specified in step S905 (step S915). For example, as in the seventeenth embodiment, the difference between the positions and angles of these two shapes (straight lines) is obtained, and the degree of coincidence is determined so as to increase as the difference decreases.

  The normality determination unit 118 corrects the error distribution obtained in step S910 according to the degree of matching obtained in step S915, and obtains a final error distribution (step S920). Specifically, in this embodiment, the median, that is, the evaluation score is corrected to be low according to the degree of matching. At this time, as the degree of matching is higher, the amount by which the evaluation score is reduced is increased.

  The normality determination unit 118 determines whether or not all edges have been selected (step S925). If it is determined that all edges have not been selected (step S925: NO), the normality determination unit 118 returns to step S905 described above. . On the other hand, when it is determined that all edges have been selected (step S925: YES), the straight line with the lowest evaluation score is determined as a white line candidate (step S930). As described above, the evaluation score is greatly reduced as the straight line has a higher degree of coincidence with the estimated far-running road shape, and therefore, it can be said that it is easily determined as a white line candidate in step S930.

  The track recognition device 100k according to the eighteenth embodiment described above has the same effect as the track recognition device 100k according to the seventeenth embodiment. In addition, among the straight lines to be evaluated in LMedS, the evaluation score is greatly reduced as the straight line having a higher degree of coincidence with the current far road shape estimated based on the far road shape calculated last time, so as a white line candidate It can be easily determined. Therefore, white line candidates can be determined with higher accuracy.

S. Other embodiments:
(S-1) In each embodiment, the process using the nearby runway shape calculated this time may be replaced with the process using the history of the nearby runway shape. For example, in step S210 of the far road shape calculation process in the first embodiment, the far road shape may be estimated using a history of the near road shape instead of the near road shape calculated this time. For example, it is also possible to estimate the far road shape from the previous near runway shape and the previous runway shape, and to combine these two far runway shapes (extrapolated) to estimate the current far road shape. Further, for example, in step S275 of the far road shape calculation process of the first embodiment, a processing region may be set using a far road shape history instead of the previously calculated far road shape.

(S-2) The fourth embodiment and the fifth embodiment may be combined. That is, when it is determined that the current position of the host vehicle is within the intersection area (step S206: YES and step S316: YES), the threshold time in step S209 and the threshold time in step S319 are both set from the normal time. May be set to a longer time.

(S-3) In each embodiment, there is no other vehicle in front of the host vehicle, the host vehicle is stopped at a red signal, and there is a moving object crossing the front of the host vehicle. In this case, a new processing area may not be set, and the processing area set before the host vehicle is stopped may be set as the current processing area. In this configuration, the road recognition device further includes an object detection unit that detects an object in front of the host vehicle, and a stop determination unit that determines whether the host vehicle is stopped with a red signal. The stop determination unit can determine whether or not the vehicle is stopped with a red signal, for example, by analyzing the captured image. When there is a moving object that crosses the front of the host vehicle, the accuracy of the nearby runway shape calculated based on the captured image is reduced, and the accuracy of the processing region that is set using such a nearby runway shape is also reduced. This lowers the accuracy of the far road shape. Therefore, by setting it as such a structure, it can suppress that the near runway shape of a low precision and a far runway shape are calculated.

(S-4) In the fifteenth embodiment, the edge of the pedestrian crossing paint used for setting the processing region is the own vehicle among the edges along the longitudinal direction of the plurality of paints identified in step S281. The edge located at the right end and the edge located at the left end in the travel lane of the present disclosure are not limited to this. It may be two edges at arbitrary positions along the longitudinal direction of the plurality of paints specified in step S281. In addition, the processing region may be specified using a plurality of edges instead of the two edges. That is, the position of the vanishing point may be specified using a plurality of edges, and the processing area may be set based on the position of the vanishing point.

(S-5) In the fifteenth embodiment, the specific method for setting the processing region can be changed as appropriate. For example, if the edge specified in step S282 is extended to specify the position of the vanishing point P10, a triangular region having a predetermined shape having the vanishing point P10 as a vertex is converted into a virtual line obtained by extending two edges. You may arrange and set along.

(S-6) In the fourth embodiment, the length of the threshold time in step S319 of the target passing point setting process is changed depending on whether or not the current position of the host vehicle is in the intersection area. The disclosure is not limited to this. For example, the step S205 of the far road shape calculation process is performed by performing a process of determining whether or not there is an input of the near road shape exceeding the predetermined time (predetermined threshold time) (hereinafter, “step S205a "). Then, steps S260 to S264 described above are executed before step S205a. At this time, the threshold time set in step S262 or S264 may be the threshold time in step S205a.

(S-7) The template image may be corrected based on at least one of the three types of parameters of the thirteenth embodiment.

(S-8) In step S800 of the far road shape calculation processing of the seventeenth embodiment, the current far road shape may be estimated using the current far road shape instead of the previous far road shape. Such estimation may be executed, for example, by the same procedure as step S210 of the far road shape calculation process of the first embodiment.

(S-9) In each embodiment, the calculated nearby runway shape and far road shape are used to set a target passing point, that is, a target passing point when the host vehicle is automatically driven. However, the present disclosure is not limited to this. For example, an alarm may be generated when at least one of the near road shape and the far road shape is not calculated, and the near road shape and the far road shape may be used as a trigger for generating such an alarm. In such a configuration, the execution of the target passing point setting unit 105 and the target passing point setting process may be omitted.

(S-10) In each embodiment, a part of the configuration realized by hardware may be replaced with software. Conversely, a part of the configuration realized by software is replaced with hardware. It may be. For example, in the first embodiment, at least one functional unit among the neighborhood shape calculation unit 101, the far shape calculation unit 102, the neighborhood shape storage unit 103, the far shape storage unit 104, and the target passing point setting unit 105 is integrated. You may implement | achieve by the module which combined the circuit, the discrete circuit, or those circuits. In addition, when part or all of the functions of the present disclosure are realized by software, the software (computer program) can be provided in a form stored in a computer-readable recording medium. “Computer-readable recording medium” is not limited to a portable recording medium such as a flexible disk or CD-ROM, but is also fixed to an internal storage device in a computer such as various types of RAM and ROM, or a computer such as a hard disk. It also includes an external storage device. That is, the “computer-readable recording medium” has a broad meaning including an arbitrary recording medium capable of fixing a data packet instead of temporarily.

  The present disclosure is not limited to the above-described embodiment, and can be realized with various configurations without departing from the spirit of the present disclosure. For example, the technical features in each embodiment are appropriately replaced or combined to solve part or all of the above-described problems or to achieve part or all of the above-described effects. It is possible. Further, if the technical feature is not described as essential in the present specification, it can be deleted as appropriate.

  100, 100a to 100k Runway recognition device, 21 imaging device, 101 near shape calculation unit, 102 far shape calculation unit, 103 near shape memory unit, 104 far shape memory unit, Er1 to Er4, El1 to El4 processing region

Claims (24)

A travel path recognition device (100, 100a to 100k) that is mounted on a vehicle and recognizes the travel path of the vehicle,
A neighborhood shape calculation unit (101) that repeatedly calculates a neighborhood track shape that is a track shape near the vehicle based on captured images repeatedly obtained by the imaging device (21) mounted on the vehicle;
A far region where a processing region (Er1 to Er4, El1 to El4) in each captured image is set based on the calculated nearby traveling route shape, and a far traveling shape which is a far traveling shape of the vehicle in the processing region is repeatedly calculated. A shape calculation unit (102);
A neighboring shape storage unit (103) that stores the calculated neighboring running road shape as a history in a time series;
A far shape storage unit (104) for storing the calculated far road shape as a history in a time series; and
With
The far-end shape calculation unit sets the processing region using at least one of the history of the near-runway shape and the history of the far-off road shape when the near-runway shape is not calculated.
Runway recognition device. In the runway recognition device according to claim 1,
A target passing point setting unit (105) for setting a target passing point used for automatic traveling of the vehicle,
The target passing point setting unit sets the target passing point based on the nearby running path shape when the neighboring running path shape is calculated, and sets the far running path shape when the neighboring running path shape is not calculated. Based on the target passing point,
Runway recognition device. In the runway recognition device according to claim 1 or 2,
The far shape calculation unit does not calculate the far track shape when the neighboring track shape is not continuously calculated beyond a predetermined first period,
Runway recognition device. In the runway recognition device according to claim 3,
If the period during which the far road shape is not calculated does not exceed a predetermined second period, the far shape calculation unit uses at least one of the near road shape history and the far road shape history. Calculating the far road shape, and if the non-calculated period exceeds the second period, do not calculate the far road shape,
Runway recognition device. In the runway recognition device according to claim 4,
A vehicle position determination unit (106) for determining whether or not the current position of the vehicle is within an intersection area including an intersection and a neighborhood area of the intersection;
As a result of the determination by the vehicle position determination unit, the distant shape calculation unit has the first period and the first period when the current position of the vehicle is within the intersection region compared to when it is not within the intersection region. Set at least one of the two periods to a longer time,
Runway recognition device. In the runway recognition device according to any one of claims 1 to 5,
An evaluation unit that acquires an evaluation value that correlates with recognition accuracy with respect to the calculated nearby runway shape, further comprising:
The evaluation unit obtains a higher evaluation value as the length in the depth direction in the calculated nearby runway shape is longer,
The distant shape calculation unit sets the processing region using the calculated nearby runway shape at a usage rate according to the calculated evaluation value.
Runway recognition device. In the runway recognition device according to claim 6,
An object detection unit (107) for detecting the presence or absence of another vehicle in front of the vehicle;
A stop determination unit (108) for determining whether or not the vehicle is stopped at a red light;
Further comprising
If the other vehicle does not exist in front of the vehicle and the vehicle is stopped at a red signal, the far-field shape calculating unit sets the processing area set before the vehicle stops, Set as the processing area this time,
Runway recognition device. In the runway recognition device according to any one of claims 1 to 6,
An object detection unit (107) for detecting an object in front of the vehicle;
A stop determination unit (108) for determining whether or not the vehicle is stopped at a red signal, using the captured image;
Further comprising
The distant shape calculation unit is a case where there is no other vehicle in front of the vehicle, the vehicle is stopped at a red signal, and there is a moving object that crosses the front of the vehicle. Does not perform a new setting of the processing area, and sets the processing area specified before the vehicle stops as the current processing area,
Runway recognition device. In the runway recognition device according to any one of claims 1 to 8,
A three-dimensional structure detection unit (109) for detecting a three-dimensional structure using a reflected wave of the irradiated electromagnetic wave;
The distant shape calculation unit identifies a roadside object existing on a side edge of a road among the detected three-dimensional structure, and sets the processing region inside an area partitioned by the identified roadside object. ,
Runway recognition device. In the runway recognition device according to any one of claims 1 to 9,
A three-dimensional structure detection unit that detects a three-dimensional structure using a reflected wave of the irradiated electromagnetic wave;
A traveling lane identifying unit (110) for identifying a current traveling lane of the vehicle;
The roadside object existing on the side edge of the road is identified among the detected three-dimensional structures, and the vehicle has passed the intersection using the identified roadside object and the identified current driving lane. A travel lane estimation unit (111) that estimates a travel lane after passing which is a lane that travels later;
Further comprising
The distant shape calculation unit sets the processing area using information indicating the estimated travel lane after passing,
Runway recognition device. In the runway recognition device according to any one of claims 1 to 10,
A posture change detector (112) for detecting a posture change of the vehicle that is nose-up or nose-down;
The distant shape calculation unit corrects and sets the position of the processing region according to the posture change when the posture change is detected.
Runway recognition device. In the runway recognition device according to any one of claims 1 to 10,
A three-dimensional structure detection unit (109) that repeatedly detects a three-dimensional structure using a reflected wave of the irradiated electromagnetic wave;
A correction amount specifying unit (114) for specifying a correction amount for correcting a position shift of the processing region in the captured image caused by a change in posture of the vehicle that is nose-up or nose-down;
Further comprising
The far shape calculation unit corrects and sets the position of the processing region with the correction amount,
The correction amount specifying unit
The difference between the distance from the vehicle to the specific point in the previous captured image and the distance to the vehicle to the specific point in the current captured image;
The difference between the distance from the vehicle to the specific three-dimensional structure detected this time by the three-dimensional structure detection unit and the difference between the distance from the vehicle to the specific three-dimensional structure detected before the previous time. Based on the correction amount,
Runway recognition device. In the runway recognition device according to any one of claims 1 to 12,
An attitude change detection unit (112) for detecting an attitude change of the vehicle that is nose-up or nose-down;
A template image specifying unit (113) for specifying, as a template image, an image of a region having a predetermined size including the processing region specified before the occurrence of the posture change;
With
The distant shape calculation unit performs pattern matching on the captured image using the template image when the posture change is detected, and sets the processing region.
Runway recognition device. In the runway recognition device according to claim 13,
The template image is based on at least one of a mounting parameter that correlates with the mounting position or orientation of the imaging device, a parameter that correlates with the shape of the traveling lane of the vehicle, and a parameter that correlates with the behavior of the vehicle. A template image correction unit for correcting is further provided.
Runway recognition device. In the runway recognition device according to claim 13 or 14,
The distant shape calculation unit gives the score indicating the degree of coincidence when the lane marking that divides the travel lane of the vehicle is a broken line, compared to the case where the lane marking is a solid line. Reduce the impact,
Runway recognition device. In the runway recognition device according to any one of claims 1 to 15,
An intersection determination unit (116) for determining whether or not the vehicle is traveling or stopping in an intersection area including an intersection and a vicinity area of the intersection;
In the captured image, a direction specifying unit (117) for detecting a plurality of paints forming a stripe pattern of a pedestrian crossing on a travel lane of the vehicle, and specifying a longitudinal direction of the detected paints;
Further comprising
The distant shape calculation unit
When it is determined that the vehicle is not running or stopped in the intersection area, the processing area is set based on the calculated nearby running road shape,
When it is determined that the vehicle is running or stopped in the intersection area, the processing area is set using the longitudinal direction of the plurality of paints specified by the direction specifying unit,
Runway recognition device. In the runway recognition device according to claim 16,
The direction specifying unit, when it is determined that the vehicle is traveling or stopped in the intersection area, of the edges along the longitudinal direction of the plurality of paints, Specify the intersection position (P10) between the virtual lines (Vl1, Vl2) extending from the edges (Eg1, Eg2),
The distant shape calculation unit sets the processing region based on the intersection position.
Runway recognition device. In the runway recognition device according to claim 17,
The two edges are located on the most end sides in the traveling lane,
Runway recognition device. In the runway recognition device according to claim 17 or claim 18,
The distant shape calculation unit includes two first points (P1r, P1l) on two imaginary lines obtained by extending two edges located at both ends in the traveling lane among edges along the longitudinal direction of the plurality of paints. And two first points that are different from the two first points on the two imaginary lines by a first distance (L21) determined in advance from the intersecting position and the two first points on the two imaginary lines. P2r, P2l), centering on the line segment connecting the two second points on the back side by a predetermined second distance (L22) from the end on the back side of the pedestrian crossing (Pd3) As the processing area, two rectangular areas (Er4, El4) having a predetermined third distance (L3) as a width in a direction intersecting the travel lane (Ln5) are specified.
Runway recognition device. In the runway recognition device according to any one of claims 1 to 19,
A normality determination unit (118) for determining the normality of the calculated far-running road shape;
The far shape calculation unit does not calculate the far road shape when it is determined that the calculated far road shape is not normal,
Runway recognition device. In the runway recognition device according to claim 20,
The distant shape calculation unit has the lowest error median frequency in the error distribution based on an error distribution obtained by LMedS based on an error between a straight line passing through any two edges in the captured image and another edge. A candidate straight line that is a straight line is calculated as a temporary far road shape,
The normality determination unit
A condition that an evaluation score, which is a frequency of a median error in the error distribution for the candidate straight line, is larger than a predetermined threshold score;
In the error distribution of the candidate straight line, a condition that a range in which an error deviating from the error median value continues is wider than a predetermined range;
A condition that the number of errors obtained is less than a predetermined threshold number;
Among these, if at least one condition is satisfied, it is determined that the temporary far road shape is not normal, and if neither condition is satisfied, the temporary far road shape is determined to be normal, Calculate the temporary far road shape as the far road shape,
Runway recognition device. In the runway recognition device according to claim 20,
Based on the history of the far road shape, further comprising a runway estimation unit (119) that estimates the estimated far road shape that is the far road shape of this time,
The distant shape calculation unit has the lowest error median frequency in the error distribution based on an error distribution obtained by LMedS based on an error between a straight line passing through any two edges in the captured image and another edge. A candidate straight line that is a straight line is calculated as a temporary far road shape,
The normality determination unit compares the estimated far road shape with the temporary far road shape, and includes an angle and a position between the estimated far road shape and the temporary far road shape. When at least one difference is equal to or greater than a predetermined difference, the provisional far road shape is determined to be not normal, and when the difference is smaller than the predetermined difference, the temporary far road shape is normal. Determining and calculating the temporary far road shape as the far road shape,
Runway recognition device. In the runway recognition device according to claim 20,
The distant shape calculation unit has the lowest error median frequency in the error distribution based on an error distribution obtained by LMedS based on an error between a straight line passing through any two edges in the captured image and another edge. A candidate straight line that is a straight line is calculated as a temporary far road shape,
The normality determination unit compares the near runway shape calculated this time with the temporary far runway shape, and compares the angle and position between the near runway shape and the temporary far runway shape. When the difference between at least one of them is greater than or equal to a predetermined difference, it is determined that the temporary far road shape is not normal, and when the difference is smaller than the predetermined difference, the temporary far road shape is normal. It is determined that there is, and the temporary far road shape is calculated as the far road shape,
Runway recognition device. In the runway recognition device according to claim 21,
Based on the history of the far road shape, further comprising a runway estimation unit (119) that estimates the estimated far road shape that is the far road shape of this time,
The normality determination unit compares a straight line passing through any two edges in the captured image with the estimated far-off road shape, and calculates an angle and a position between the straight line and the estimated far-off road shape. An evaluation score that is the frequency of the median error in the error distribution for the straight line is set to be smaller as the difference between at least one of
Runway recognition device.

Images