JP2018041396A - Object recognition device, object recognition method, and object recognition program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide one of the objects which is to provide an object recognition device, an object recognition method, and an object recognition program capable of suppressing an excessive exclusion of an area when limiting the area when preferentially performing the process by limiting the area in which the object is presumed to exist.SOLUTION: An object recognition device includes: a detection part for detecting a target of the surrounding area of a vehicle; a recognition part for recognizing an object based on the detected target; and a processing object setting part for excluding targets of the target group from the recognition object when the number of targets of the target group including g one or a plurality of the detected targets along a vehicle travel direction from among the detected targets. The processing object setting part does not exclude the targets of the target group, but defines them as the recognition objects when the number of the targets of the target group is equal to or more than the predetermined number and further when a degree of the fluctuation of the targets of the target group is equal to or less than a threshold value.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an object recognition device, an object recognition method, and an object recognition program.

  Conventionally, in image processing, an edge (characteristic portion) is extracted from an image captured by a camera in order to specify an object. In such image processing, a technique for shortening the time required for object recognition by excluding the area for object recognition from the image based on the number of candidate image portions where there may be extracted pedestrians is known. (For example, refer to Patent Document 1).

JP-A-2006-12303

  However, in the conventional technique, there is a case where a region where an object to be recognized originally exists is temporarily removed from an image. As a result, vehicle control over the object is delayed, and merchantability may be deteriorated.

  The present invention has been made in consideration of such circumstances, and when processing is preferentially focused on a region where an object is likely to exist, the region is excessively excluded when narrowing the region. An object of the present invention is to provide an object recognition device, an object recognition method, and an object recognition program that can suppress this.

  The invention according to claim 1 is a detection unit for detecting a target around a vehicle, a recognition unit for recognizing an object based on the target detected by the detection unit, and a target detected by the detection unit. If the number of targets in the target group including one or more targets detected along the traveling direction of the vehicle is a predetermined number or more, the target included in the target group is A processing target setting unit that is excluded from recognition targets by a recognition unit, and the processing target setting unit further includes a target in the target group when the number of targets in the target group is equal to or greater than a predetermined number. When the degree of variation is less than the threshold value, the target recognition device is configured to recognize a target included in the target group without excluding it from the recognition target by the recognition unit.

  According to a second aspect of the present invention, in the object recognition device according to the first aspect, the processing target setting unit includes the target group when the degree of variation of the target in the target group is equal to or less than a threshold value. The target included in the target group is continuously recognized as a recognition target by the recognition unit until the vehicle passes through the position of the target included.

  According to a third aspect of the present invention, there is provided a camera that images the front of a vehicle, an extraction unit that extracts a feature location that can be recognized as an object from an image captured by the camera, and a feature location that is extracted by the extraction unit. A recognition unit for recognizing an object on the basis of the image, and the number of feature locations in the feature location group including one or more feature locations extracted along the depth direction of the image by the extraction unit is a predetermined number or more. A processing target setting unit configured to exclude a feature location included in the feature location group from a recognition target by the recognition unit, and the processing target setting unit includes a number of feature locations included in the feature location group. Is greater than or equal to a predetermined number, and if the degree of variation of the feature locations included in the feature location group is less than a threshold, the features included in the feature location group Location and be recognized without excluding from the recognition target by the recognition unit, a object recognition apparatus.

  In the invention according to claim 4, the in-vehicle computer detects a target around the vehicle, recognizes the detected target as an object, and detects the detected target along the traveling direction of the vehicle. When the number of targets in the target group including one or a plurality of targets is equal to or greater than a predetermined number, the targets included in the target group are excluded from recognition targets, and the targets in the target group When the number of target objects is greater than or equal to a predetermined number and the degree of variation of the target in the target group is less than a threshold, the target included in the target group is not excluded from the recognition target. This is an object recognition method.

  The invention according to claim 5 causes the in-vehicle computer to detect a target around the vehicle, recognizes the detected target as an object, and follows the traveling direction of the vehicle among the detected targets. When the number of targets in the target group including one or more targets detected in the above is a predetermined number or more, the target included in the target group is excluded from recognition targets, and the target group When the number of targets is equal to or greater than a predetermined number, and the degree of variation of the targets in the target group is less than a threshold value, the targets included in the target group are not excluded from recognition targets. An object recognition program to be recognized.

  According to the first, third, fourth, and fifth aspects of the invention, when the number of targets in the target group including one or more targets detected along the traveling direction of the vehicle is equal to or greater than a predetermined number. Furthermore, if the degree of target variation in the target group is less than the threshold value, the target process included in the target group is not excluded from the recognition targets, but limited processing is performed. It is possible to suppress excessive exclusion of a region when narrowing down the region when preferentially performing processing while narrowing down to a region where an object (such as a pedestrian) is likely to exist in time. .

  According to the invention described in claim 2, when the degree of variation of the target in the target group is equal to or less than the threshold, the period until the vehicle passes through the position of the target included in the target group, the target group The accuracy of object recognition can be improved by keeping the target included in the object as a recognition target at the time of object recognition.

It is a figure which shows an example of a structure of the object recognition apparatus 100 in embodiment. 4 is a flowchart illustrating an example of a flow of processing by a control unit 110. It is a figure which shows typically a mode that the candidate image part 134 is extracted from the captured image 132. FIG. It is a figure which shows an example of the coordinate system by which the position of the candidate image part 134 extracted from the captured image 132 shown in FIG. 3 was mapped. 6 is a diagram illustrating an example of a distribution of candidate image portions 134 in a Y-axis direction of a determination region R. FIG. It is a figure which shows typically a mode that the pedestrian who exists between 2 ridges is extracted as the candidate image part. It is a figure which shows an example of the coordinate system by which the position of the candidate image part 134 extracted from the captured image 132 shown in FIG. 6 was mapped. 12 is a diagram illustrating another example of the distribution of candidate image portions 134 in the Y-axis direction of the determination region R. FIG. It is a figure which shows an example of the information 140 for time limit setting. It is a figure which shows typically a mode that the candidate image part 134 is excluded from a process target according to dispersion | distribution of the candidate image part 134. FIG.

  Hereinafter, embodiments of an object recognition device, an object recognition method, and an object recognition program of the present invention will be described with reference to the drawings.

  FIG. 1 is a diagram illustrating an example of a configuration of an object recognition apparatus 100 according to the embodiment. The object recognition apparatus 100 according to the present embodiment is mounted on a vehicle (hereinafter referred to as own vehicle M). The object recognition device 100 includes, for example, a camera 102, a control unit 110, and a storage unit 130. The object recognition apparatus 100 is realized by, for example, one or more processors or hardware having an equivalent function. The object recognition apparatus 100 is configured by combining a processor such as a CPU (Central Processing Unit), a storage device, and an ECU (Electronic Control Unit) in which a communication interface is connected by an internal bus, or an MPU (Micro-Processing Unit). It may be. Further, the object recognition device 100 may be distributed by a plurality of computer devices.

  The camera 102 is a digital camera using a solid-state image sensor such as a charge coupled device (CCD) or a complementary metal oxide semiconductor (CMOS). The camera 102 is provided, for example, on the upper part of the front windshield in the passenger compartment or on the rear surface of the room mirror. For example, the camera 102 periodically images the front of the host vehicle M repeatedly. The camera 102 may be a stereo camera including a plurality of cameras.

  For example, the control unit 110 includes an image acquisition unit 112, a candidate image portion extraction unit 114, a distance calculation unit 116, a processing target setting unit 118, a candidate image portion determination unit 120, a pedestrian image identification unit 122, A pedestrian orientation determination unit 124, an avoidance determination unit 126, and an alarm output unit 128 are provided. These components are realized by a processor executing a program (software). Some or all of these components may be realized by hardware such as LSI (Large Scale Integration) and ASIC (Application Specific Integrated Circuit), or may be realized by a combination of software and hardware. Also good.

  The storage unit 130 is realized by a ROM (Read Only Memory), a RAM (Random Access Memory), an HDD (Hard Disk Drive), a flash memory, or the like. The program executed by the processor may be stored in advance in the storage unit 130, or may be downloaded from an external device via an in-vehicle Internet facility or the like. Further, the program may be installed in the storage unit 130 by attaching a portable storage medium storing the program to a drive device (not shown). The storage unit 130 stores, for example, information such as the captured image 132, the candidate image portion 134, the simple matching data 136, the pedestrian matching data 138, and the time limit setting information 140.

  Hereinafter, each component with which the control part 110 is provided is demonstrated using a flowchart. FIG. 2 is a flowchart illustrating an example of a processing flow by the control unit 110.

  First, the image acquisition unit 112 acquires an image from the camera 102 (step S <b> 100) and stores it as a captured image 132 in the storage unit 130. The image acquired from the camera 102 may be, for example, a color image or a gray scale image (luminance image).

  The image acquisition unit 112 is captured by the external camera by wirelessly communicating with an external camera installed on the side of the road or above the road instead of acquiring the captured image 132 from the camera 102 mounted on the host vehicle M. The captured image may be acquired as the captured image 132. The wireless communication may be performed by using, for example, a cellular network, a Wi-Fi network, Bluetooth (registered trademark), DSRC (Dedicated Short Range Communication), or the like. In this case, in addition to (or instead of) the camera 102 mounted on the host vehicle M, the camera in the claims may include an external camera installed on the side of the road or above the road.

  Next, the candidate image part extraction unit 114 extracts feature points from all or a part of the captured image 132, and compares the extracted feature points with the simple matching data 136 stored in the storage unit 130 in advance. Thus, a region where a pedestrian may exist (a region including a collection of feature points matching the simple matching data 136) is extracted as a candidate image portion 134 (step S102). The candidate image portion 134 is an example of a “target”. The camera 102, the image acquisition unit 112, and the candidate image part extraction unit 114 are examples of “detection unit”.

  The “feature portion” is a feature point having a luminance difference or color parameter difference from an edge, that is, a surrounding pixel, or a pixel group having an equivalent property.

  The simple matching data 136 is data created based on a gentler standard than the pedestrian matching data 138. “Loose reference” means, for example, that shape model data described later is represented by fewer characteristic portions than the shape model data included in the pedestrian matching data 138. “Loose criteria” may mean that the required level of “match” is lower in the process of step S102 than in the process of step S124 described later. For example, the processing in step S102 may be such that the number of shape models that are required to match is lower than in step S124.

  The pedestrian matching data 138 is, for example, shape model data indicating a part of a pedestrian's body (eg, head, shoulders, hands, feet, eyes, nose, mouth, etc.) or the entire outline or ridgeline. The pedestrian matching data 138 may include a plurality of shape model data corresponding to a plurality of parts (parts) of the pedestrian's body. On the other hand, the simple matching data 136 is simpler shape model data than the shape model data included in the pedestrian matching data 138, and is data that is easily determined to match the feature location group to be compared.

  The candidate image portion extraction unit 114 extracts a region where the extracted feature location indicates a shape that matches the shape model data included in the simple matching data 136 as the candidate image portion 134 and causes the storage unit 130 to store the extracted region. In this process, since the feature locations indicating the shape features are compared with each other, a portion showing an object other than a pedestrian (human) may be extracted as the candidate image portion 134.

  FIG. 3 is a diagram schematically illustrating how the candidate image portion 134 is extracted from the captured image 132. In the illustrated example, after two wheel ridges are formed on the snow road, the snow melts only in the ridge portion, and a snow mass remains between the two ridges, and this snow mass is extracted as a candidate image portion 134. It has been.

  The distance calculation unit 116 calculates the distance from the host vehicle M for each candidate image portion 134 extracted by the candidate image portion extraction unit 114 (step S104). For example, the distance calculation unit 116 derives a vanishing point of a road that is an infinite point in the captured image 132 and derives a distance between pixels of the vanishing point and the candidate image portion 134. Then, the distance calculation unit 116 calculates an angle (elevation angle) with respect to the infinity point based on the focal length of the camera 102 and the distance between the vanishing point and the pixel of the candidate image portion 134, and the angle and the camera 102 are installed. The distance in the image depth direction between each candidate image portion 134 and the host vehicle M is obtained by obtaining a position in the real space corresponding to the position of the candidate image portion 134 in the captured image 132 based on the height of the position thus obtained. calculate.

  The candidate image portion determination unit 120 maps the position of the candidate image portion 134 for which the distance is calculated by the distance calculation unit 116 to a coordinate system indicating the real space. The coordinate system indicating the real space may be represented, for example, in a two-dimensional space when the host vehicle M is viewed from above.

FIG. 4 is a diagram illustrating an example of a coordinate system in which the position of the candidate image portion 134 extracted from the captured image 132 illustrated in FIG. 3 is mapped. In the figure, the X axis represents the horizontal direction, and the Y axis represents the depth direction. In addition, O in the drawing represents, for example, a position from the front end position of the host vehicle M (for example, a position such as a front bumper). In the figure, R _C represents a determination area set for a travel lane in which the host vehicle M travels, and R _LA and R _LB are left lanes (or areas outside the road) with respect to the traveling direction of the host vehicle M. ) Represents the determination area set in (). R _RA and R _RB represent determination areas set in the right lane (or the area outside the road) with respect to the traveling direction of the host vehicle M.

  For example, the candidate image portion determination unit 120 converts the distance from the host vehicle M to the candidate image portion 134 into coordinates on the Y axis, and converts the position of the horizontal component of the candidate image portion 134 in the captured image 132 into coordinates on the X axis. To map the position of each candidate image portion 134 to the XY coordinate system.

Next, the candidate image portion determination unit 120 has a width in the horizontal direction (X-axis direction) orthogonal to the traveling direction (Y-axis direction) of the host vehicle M in the coordinate system (XY coordinate system) indicating the real space. A determination region R that is within a predetermined width is set. For example, the candidate image portion determination unit 120 sets a plurality of determination regions R such as R _C , R _LA , R _LB , R _RA , R _{RB in} FIG. Then, the candidate image portion determination unit 120 selects one determination region of interest from the set determination regions R (step S106). When a plurality of determination regions R are set, the predetermined width in the horizontal direction of each determination region R may be the same or different. Moreover, each determination area | region R may overlap with the adjacent determination area | region.

Next, the candidate image portion determination unit 120 counts the number n of candidate image portions 134 belonging to the determination region R selected in S106 among the mapped candidate image portions 134. In the example of FIG. 4, when the determination region _RC is selected from the plurality of determination regions R, the candidate image portion 134a is a candidate image portion that does not belong to the determination region _RC , and the candidate image portion 134b is the determination region R. It becomes a candidate image portion belonging to _C. Then, the candidate image portion determination unit 120 determines whether or not the number n of candidate image portions 134 (134b in the figure) belonging to the determination region R is equal to or greater than a predetermined number k (step S108). The candidate image portion belonging to the determination region R is an example of a “target group” or “characteristic location group”.

  When the candidate image portion determination unit 120 determines that the number n of candidate image portions 134 belonging to the determination region R is less than the predetermined number k, the processing target setting unit 118 determines that the candidate image portion determination unit 120 has selected. All candidate image portions 134 belonging to the region R are set as processing targets of the pedestrian image identification unit 122 described later (step S110).

  On the other hand, the processing target setting unit 118 is selected by the candidate image portion determination unit 120 when the candidate image portion determination unit 120 determines that the number n of candidate image portions 134 belonging to the determination region R is a predetermined number k or more. The variance of the candidate image portion 134 belonging to the determination region R is derived. The distribution of the candidate image portion 134 is an example of “the degree of variation in the target” or “the degree of variation in the characteristic portion”. Then, the processing target setting unit 118 determines whether or not the derived variance is less than a threshold value (step S112).

  FIG. 5 is a diagram illustrating an example of the distribution of the candidate image portion 134 in the Y-axis direction of the determination region R. In the drawing, the vertical axis represents the number of candidate image portions 134 (number of candidates n), and the horizontal axis represents the Y-axis direction of the determination region R. For example, since a lump of snow or the like existing between the two fences is likely to be omnipresent along the roadway, as shown in FIG. 5, the candidate image portion 134 is mapped with a certain degree of dispersion in the Y-axis direction. There is a tendency to. In this case, the variance of the candidate image portion 134 tends to increase. That is, the variance of the candidate image portion 134 tends to exceed the threshold value.

  The process target setting unit 118 sets the elapsed time threshold Tth to the predetermined time T when the variance of the candidate image portion 134 is equal to or greater than the threshold (step S114). The elapsed time threshold value Tth is a time period for prohibiting (not shifting) the processing of S120 described later until a time corresponding to the threshold value elapses.

  On the other hand, when there is a pedestrian between the two ridges, it is assumed that the pedestrian is extracted as the candidate image portion 134 together with the unmelted snow in the captured image 132. FIG. 6 is a diagram schematically illustrating how a pedestrian existing between two ridges is extracted as a candidate image portion 134. As shown in FIG. 6, the feature location of the pedestrian is more easily matched with the shape data included in the simple matching data 136 than the feature location of the other object, so that more candidates are present at the location where the pedestrian exists. There is a tendency for the image portion 134 to be extracted. That is, a point where the density of candidate image portions 134 (the number n of candidate image portions 134 per predetermined section in the Y-axis direction) exceeds a threshold value can be evaluated as having a high possibility of a pedestrian.

  FIG. 7 is a diagram illustrating an example of a coordinate system in which the position of the candidate image portion 134 extracted from the captured image 132 illustrated in FIG. 6 is mapped. For example, a large number of candidate image portions 134 are extracted in the vicinity of the coordinate A corresponding to the point where the pedestrian is present in the captured image 132 as compared with other coordinates.

  FIG. 8 is a diagram illustrating another example of the distribution of the candidate image portion 134 in the Y-axis direction of the determination region R. The distribution shown in FIG. 8 corresponds to the candidate image portion 134 mapped in FIG. As shown in the figure, when a pedestrian exists in the selected determination region R, the number of candidate image portions 134 tends to protrude at a certain point in the Y-axis direction, and the dispersion of the candidate image portions 134 tends to be small. That is, when a pedestrian is present in the determination region R, the variance of the candidate image portion 134 tends to be less than the threshold value.

  If the process target setting unit 118 determines in step S112 that the variance of the candidate image portion 134 is less than the threshold, the process target setting unit 118 sets the elapsed time threshold Tth to a time T + α that is an extension of the predetermined time T (step S116). For the extended time α, the elapsed time threshold Tth is extended to a time at which the host vehicle M can pass the point where the variance of the candidate image portion 134 exceeds the threshold (the point where the candidate image portions 134 are densely packed). . The extended time α may be, for example, a predetermined time, or decreases as the vehicle speed of the host vehicle M increases from a certain reference time, and the vehicle speed of the host vehicle M decreases. The time may be increased accordingly.

  For example, when the variance of the candidate image portion 134 is less than the threshold value, it can be determined that there is a high possibility that a pedestrian is present in the determination region R. Therefore, a recognition process performed by the pedestrian image identification unit 122 described below. Until it is recognized whether or not there is a pedestrian at a point where the number n of candidate image portions 134 protrudes (that is, a point where there is a high possibility that a pedestrian is present). It is preferable to continue to recognize whether the candidate image portion 134 is a pedestrian without continuing to exclude the image portion 134 from the processing target uniformly. Accordingly, the candidate image portion 134 is a pedestrian without excluding the candidate image portion 134 in the determination region R from the processing target until the predetermined time T or the time T + α obtained by extending the predetermined time T elapses. It is necessary to continue to recognize whether or not.

  For example, the processing target setting unit 118 refers to the time limit setting information 140 stored in the storage unit 130, sets the elapsed time threshold Tth according to the speed V of the host vehicle M, and starts counting time. . FIG. 9 is a diagram illustrating an example of the time limit setting information 140. The time limit setting information 140 is information indicating the relationship between the elapsed time threshold value Tth and the speed V of the host vehicle M, and is represented by, for example, a table, a map, a mathematical expression, or the like. As illustrated, the elapsed time threshold value Tth has a tendency to decrease as the speed V of the host vehicle M increases. Therefore, the process target setting unit 118 sets the elapsed time threshold Tth to be shorter as the speed V of the host vehicle M is higher.

  Next, the processing target setting unit 118 determines whether or not the elapsed time t is equal to or greater than the elapsed time threshold Tth set to T or T + α (step S118). The elapsed time t is, for example, the time elapsed from the time when the processing of this flowchart is started to the current time.

  When the elapsed time t is equal to or greater than the elapsed time threshold Tth, the processing target setting unit 118 determines that the candidate image portion 134 extracted in the determination region R is unlikely to be a region where a pedestrian exists. Then, the candidate image portion 134 in the determination region R selected by the candidate image portion determination unit 120 is excluded from the processing target (step S120). As a result, the candidate image portion 134 included in the determination region R selected by the candidate image portion determination unit 120 is excluded from the processing target of the pedestrian image identification unit 122 described later. Note that the processing target setting unit 118 may exclude all candidate image portions 134 included in the determination region R from the processing target, or may exclude some candidate image portions 134 from the processing target. Some candidate image portions 134 are, for example, candidate image portions 134 close to the host vehicle M in the determination region R. Thereby, for example, even if the candidate image portion 134 to be excluded is included in the processing target of the pedestrian image identification unit 122 when the distance from the host vehicle M is long and the processing time is sufficient. .

  On the other hand, when the elapsed time t is less than the elapsed time threshold Tth, the processing target setting unit 118 proceeds to the process of S110 described above, and all candidate image parts belonging to the determination region R selected by the candidate image part determination unit 120 134 is set as a processing target of the pedestrian image identification unit 122. Thus, until the elapsed time threshold value Tth elapses, the candidate image portion 134 in the determination region R is not excluded from the processing target, and it is continuously recognized whether the candidate image portion 134 is a pedestrian. Become.

  In the above-described processing, the predetermined time T and the time T + α obtained by extending the predetermined time T are used as the elapsed time threshold Tth. However, the present invention is not limited to this, and the predetermined time without using the extended time + α. Only T may be used. Specifically, when the variance of the candidate image portion 134 is less than the threshold in the processing of S112, the processing of S114 and S118 is skipped and the processing of S120 is performed, and the variance of the candidate image portion 134 is greater than or equal to the threshold. The elapsed time threshold value Tth may be set to the predetermined time T in S116, and it may be determined whether or not the elapsed time threshold value Tth has elapsed in the process of S118.

  Next, the processing target setting unit 118 determines whether or not the candidate image portion 134 belonging to the determination region R is excluded from the processing target or the processing target processing has been performed for all the set determination regions R ( Step S122). When the processing target setting unit 118 does not perform processing for excluding or setting candidate image portions 134 belonging to the determination region R as processing targets for all the determination regions R, the processing target setting unit 118 moves the process to S106 described above. A determination area R that has not yet been selected is selected as a determination area of interest.

  On the other hand, when the processing is performed for all the determination regions R, the pedestrian image identification unit 122 uses the pedestrian matching data 138 to set the candidate image portion 134 to be processed set by the processing target setting unit 118. , A pedestrian identification process for identifying whether or not the image includes a pedestrian image portion is performed (step S124). The image part of the pedestrian is an area on the image that can be determined that the object that can be estimated based on the characteristic part included in the image part represents the pedestrian.

  For example, the pedestrian image identification unit 122 selects a candidate image portion 134 having a characteristic location that matches the shape data included in the pedestrian matching data 138 from the candidate image portions 134 to be processed. In other words, the pedestrian image identification unit 122 is more detailed than the simple matching data 136 referred to in the processing of the candidate image part extraction unit 114 (for example, the number of feature points to be matched is several times larger). By using 138, the candidate image portion 134 including the pedestrian image portion is selected from the candidate image portions 134 to be processed. The pedestrian image identification unit 122 and the pedestrian orientation determination unit 124 are examples of a “recognition unit”.

  The pedestrian orientation determination unit 124 determines the pedestrian orientation of each candidate image portion 134 selected by the pedestrian image identification unit 122 (step S126). For example, the pedestrian orientation determination unit 124 derives the pedestrian's posture, or the orientation of the head or limbs from the pedestrian's image portion, and determines the pedestrian's orientation based on the posture or the orientation of the body part. Determine. When the candidate image portion 134 including the pedestrian image portion is not selected by the pedestrian image identification unit 122 among the plurality of candidate image portions 134 to be processed, the processing of this flowchart ends.

  The avoidance determination unit 126 determines whether or not the pedestrian is to be avoided according to the direction of the pedestrian determined by the pedestrian direction determination unit 124 (step S128). For example, the avoidance determination unit 126 avoids the pedestrian when the direction of the pedestrian is a direction approaching the own vehicle M or when the pedestrian is stationary in the center of the traveling lane on which the own vehicle M is traveling. Judge as an object.

  When the avoidance determination unit 126 determines that the pedestrian is an avoidance target, the alarm output unit 128 outputs an alarm that prompts the driver to avoid using the in-vehicle speaker or the display device (step S130). Thereby, the process of this flowchart is complete | finished.

  FIG. 10 is a diagram schematically illustrating a state in which the candidate image portion 134 is excluded from the processing target according to the distribution of the candidate image portion 134. In the figure, (a) represents a scene where the variance of the candidate image portion 134 is greater than or equal to the threshold, and (b) represents a scene where the variance of the candidate image portion 134 is less than the threshold. For example, in the scene of (a), after the point P1 at which time equal to or greater than the elapsed time threshold Tth has elapsed since it was determined that the number n of candidate image portions 134 belonging to the determination region R is a predetermined number k or more, the determination region The candidate image portion 134 belonging to R is excluded from the processing target. As a result, the number of candidate image portions 134 to be subjected to pedestrian recognition processing is reduced, so that the time required for pedestrian recognition can be shortened. Moreover, in the scene of (b), since pedestrian recognition is continued to the point P2 which passed the point A with a high possibility that a pedestrian exists to some extent, the own vehicle M can be drive | worked more safely.

  In (a) and (b), points P1 and P2 and subsequent points are excluded in all excluded areas, but it is actually determined whether the elapsed time after exceeding a predetermined number k for each excluded area exceeds the elapsed time threshold value Tth. Therefore, the exclusion timing is different for each household area.

  Further, in (b), the area after the point A is excluded, but the area from P1 to A may also be excluded. In this case, instead of processing using the elapsed time threshold value Tth, control is performed such that the determination region R is divided into a plurality of regions in the distance direction, and only the region corresponding to the portion where the candidate image portions 134 are concentrated is set as a control target. . Note that this processing is not limited to use for excluding the region from P1 to A, but can be used instead of the processing using the elapsed time threshold value Tth.

  According to the object recognition device 100 in the embodiment described above, when the number n of candidate image portions 134 belonging to the determination region R is less than a predetermined number k, the candidate image portions 134 belonging to the determination region R are set as processing targets. When the number n of candidate image portions 134 belonging to the determination region R is equal to or greater than a predetermined number k, if the variance of the candidate image portions 134 belonging to the determination region R is equal to or greater than a threshold, the candidate image portions 134 belonging to the determination region R are When the variance of the candidate image portion 134 belonging to the determination region R is excluded from the processing target and the variance of the candidate image portion 134 belonging to the determination region R is less than the threshold value, the candidate image portion 134 belonging to the determination region R is not excluded from the processing target and is set as the recognition target. When processing is preferentially performed by narrowing down a region that will exist, it is possible to suppress excessive exclusion of the region when narrowing down the region.

  Further, according to the object recognition device 100 in the above-described embodiment, when the variance of the candidate image portion 134 belonging to the determination region R is less than the threshold, the host vehicle M passes through a point where there is a high possibility that a pedestrian exists. By continuing the candidate image portion 134 belonging to the determination region R as a processing target until the period until the determination is made, the accuracy in object recognition can be improved, and the host vehicle M can be driven more safely.

  As mentioned above, although the form for implementing this invention was demonstrated using embodiment, this invention is not limited to such embodiment at all, In the range which does not deviate from the summary of this invention, various deformation | transformation and substitution Can be added.

DESCRIPTION OF SYMBOLS 100 ... Object recognition apparatus, 102 ... Camera, 110 ... Control part, 112 ... Image acquisition part, 114 ... Candidate image part extraction part, 116 ... Distance calculation part, 118 ... Processing target setting part, 120 ... Candidate image part determination part, 122: Pedestrian image identification unit, 124 ... Pedestrian orientation determination unit, 126 ... Avoidance determination unit, 128 ... Alarm output unit, 130 ... Storage unit, 132 ... Captured image, 134 ... Candidate image portion, 136 ... Simple matching data, 138 ... Pedestrian matching data, 140 ... Time limit setting information

Claims (5)

A detection unit for detecting a target around the vehicle;
A recognition unit for recognizing an object based on the target detected by the detection unit;
Among the targets detected by the detection unit, when the number of targets in a target group including one or more targets detected along the traveling direction of the vehicle is equal to or greater than a predetermined number, A processing target setting unit that excludes targets included in the target group from recognition targets by the recognition unit,
The processing target setting unit
When the number of targets in the target group is equal to or greater than a predetermined number, and when the degree of variation in the target group is less than a threshold, the target included in the target group is recognized as the recognition target. To be recognized without being excluded from recognition by
Object recognition device. The processing target setting unit includes a period of time until the vehicle passes the position of the target included in the target group when the degree of variation of the target in the target group is a threshold value or less. Continue to include targets included in the group as recognition targets by the recognition unit.
The object recognition apparatus according to claim 1. A camera that images the front of the vehicle;
An extraction unit that extracts a feature location that can be recognized as an object from an image captured by the camera;
A recognition unit for recognizing an object on the image based on the feature location extracted by the extraction unit;
Features included in the feature location group when the number of feature locations included in the feature location group including one or more feature locations extracted along the depth direction of the image by the extraction unit is a predetermined number or more. A processing target setting unit that excludes a part from a recognition target by the recognition unit,
The processing target setting unit
When the number of feature locations included in the feature location group is equal to or greater than a predetermined number, and further when the degree of variation of the feature locations included in the feature location group is less than a threshold value, the feature locations included in the feature location group , As a recognition target without excluding from the recognition target by the recognition unit,
Object recognition device. In-vehicle computer
Detect targets around the vehicle,
Recognizing the detected target as an object,
Among the detected targets, when the number of targets in the target group including one or more targets detected along the traveling direction of the vehicle is a predetermined number or more, included in the target group Target to be recognized,
When the number of targets in the target group is equal to or greater than a predetermined number, and the degree of variation of the target in the target group is less than a threshold, the targets included in the target group are recognized To be recognized without being excluded from
Object recognition method. On-board computer
Detect targets around the vehicle,
Recognizing the detected target as an object,
Among the detected targets, when the number of targets in the target group including one or more targets detected along the traveling direction of the vehicle is a predetermined number or more, the target group Excluding included targets from recognition,
When the number of targets in the target group is equal to or greater than a predetermined number, and the degree of variation of the target in the target group is less than a threshold, the targets included in the target group are recognized Make it a recognition target without excluding it from
Object recognition program.

Images