JP5999183B2 - Three-dimensional object detection apparatus and three-dimensional object detection method - Google Patents

Description

The present invention relates to a three-dimensional object detection device and a three-dimensional object detection method.
This application claims priority based on Japanese Patent Application No. 2012-166520 filed on Jul. 27, 2012. For designated countries that are allowed to be incorporated by reference, The contents described in the application are incorporated into the present application by reference and made a part of the description of the present application.

  2. Description of the Related Art Conventionally, there is known a technique for converting two captured images captured at different times into a bird's-eye view image and detecting a three-dimensional object based on the difference between the two converted bird's-eye view images (Patent Literature). 1).

JP 2008-227646 A

  When a three-dimensional object existing in the adjacent lane adjacent to the traveling lane of the host vehicle is detected as an adjacent vehicle based on the captured image captured by the camera, foreign matter such as scale adheres to the lens, and the lens becomes cloudy. (A thin white film is formed on the lens surface), part of the light beam from the subject is blocked or diffusely reflected by foreign matter adhering to the lens. As a result, there is a case where the adjacent vehicle cannot be detected properly.

  The problem to be solved by the present invention is to provide a three-dimensional object detection device that can appropriately detect an adjacent vehicle even when foreign matters such as water scale adhere to the lens and the lens is clouded.

  The present invention detects a light source existing behind the host vehicle and calculates a certainty factor that the detected light source is a headlight of an adjacent vehicle. Further, when calculating the reliability of the headlight of the adjacent vehicle, the white turbidity of the lens is calculated based on the captured image, and the higher the turbidity of the lens, the higher the reliability of the headlight of the adjacent vehicle. And the said subject is solved by accelerating | stimulating detecting a solid object, so that the certainty degree of the headlight of an adjacent vehicle is high.

  According to the present invention, when the lens is clouded, by calculating a high certainty that the detected light source is the headlight of the adjacent vehicle, based on the certainty of the headlight of the adjacent vehicle, The headlight of the adjacent vehicle can be detected appropriately, so that the adjacent vehicle can be detected appropriately even when the lens is clouded.

It is a schematic block diagram of the vehicle carrying a solid-object detection apparatus. It is a top view which shows the driving state of the vehicle of FIG. It is a block diagram which shows the detail of a computer. It is a figure for demonstrating the outline | summary of the process of a position alignment part, (a) is a top view which shows the movement state of a vehicle, (b) is an image which shows the outline | summary of position alignment. It is the schematic which shows the mode of the production | generation of the difference waveform by a solid-object detection part. It is a figure which shows an example of threshold value (alpha) for detecting a differential waveform and a solid object. It is a figure which shows the small area | region divided | segmented by the solid-object detection part. It is a figure which shows an example of the histogram obtained by a solid-object detection part. It is a figure which shows the weighting by a solid-object detection part. It is a figure which shows the other example of the histogram obtained by a solid-object detection part. It is a figure for demonstrating the relationship between the cloudiness of a lens, and the sharpness of an image. It is a figure for demonstrating an example of the method of calculating the sharpness of an image. It is a figure for demonstrating the relationship between the certainty degree of the headlight of an adjacent vehicle, and a difference threshold value. It is a flowchart which shows the adjacent vehicle detection process which concerns on 1st Embodiment. It is a flowchart which shows the threshold value change process which concerns on 1st Embodiment. It is a block diagram which shows the detail of the computer which concerns on 2nd Embodiment. It is a figure which shows the driving | running | working state of a vehicle, (a) is a top view which shows positional relationships, such as a detection area, (b) is a perspective view which shows positional relationships, such as a detection area in real space. It is a figure for demonstrating operation | movement of the brightness | luminance difference calculation part which concerns on 2nd Embodiment, (a) is a figure which shows the positional relationship of the attention line, reference line, attention point, and reference point in a bird's-eye view image, (b). FIG. 4 is a diagram illustrating a positional relationship among attention lines, reference lines, attention points, and reference points in real space. It is a figure for demonstrating the detailed operation | movement of the brightness | luminance difference calculation part which concerns on 2nd Embodiment, (a) is a figure which shows the detection area | region in a bird's-eye view image, (b) is the attention line and reference line in a bird's-eye view image. It is a figure which shows the positional relationship of an attention point and a reference point. It is a figure which shows the example of an image for demonstrating edge detection operation | movement. It is a figure which shows the luminance distribution on an edge line and an edge line, (a) is a figure which shows luminance distribution in case a solid object (adjacent vehicle) exists in a detection area, (b) is a figure which shows a solid object in a detection area. It is a figure which shows the luminance distribution when it does not exist. It is a flowchart which shows the adjacent vehicle detection method which concerns on 2nd Embodiment. It is a flowchart which shows the threshold value change process which concerns on 2nd Embodiment.

<< First Embodiment >>
FIG. 1 is a schematic configuration diagram of a vehicle equipped with a three-dimensional object detection device 1 according to the present embodiment. The three-dimensional object detection device 1 according to the present embodiment is intended to detect another vehicle (hereinafter also referred to as an adjacent vehicle V2) existing in an adjacent lane that may be contacted when the host vehicle V1 changes lanes. To do. As shown in FIG. 1, the three-dimensional object detection device 1 according to the present embodiment includes a camera 10, a vehicle speed sensor 20, and a calculator 30.

  As shown in FIG. 1, the camera 10 is attached to the vehicle V <b> 1 so that the optical axis is at an angle θ from the horizontal to the lower side at the position of the height h behind the host vehicle V <b> 1. The camera 10 captures an image of a predetermined area in the surrounding environment of the host vehicle V1 from this position. The vehicle speed sensor 20 detects the traveling speed of the host vehicle V1, and calculates the vehicle speed from the wheel speed detected by, for example, a wheel speed sensor that detects the rotational speed of the wheel. The computer 30 detects an adjacent vehicle existing in an adjacent lane behind the host vehicle.

  FIG. 2 is a plan view showing a traveling state of the host vehicle V1 of FIG. As shown in the figure, the camera 10 images the rear of the vehicle at a predetermined angle of view a. At this time, the angle of view a of the camera 10 is set to an angle of view at which the left and right lanes (adjacent lanes) can be imaged in addition to the lane in which the host vehicle V1 travels. The area that can be imaged includes the detection target areas A1 and A2 on the adjacent lane on the left and right sides of the traveling lane of the host vehicle V1 behind the host vehicle V1. Note that the rear of the vehicle in this embodiment includes not only the rear of the vehicle but also the side of the rear of the vehicle. The area behind the imaged vehicle is set according to the angle of view of the camera 10. Although it is an example, when the right rear of the vehicle along the vehicle length direction is set to zero degrees, it can be set so as to include a region of 0 degrees to 90 degrees, preferably 0 degrees to 70 degrees on the left and right sides from the right rear direction.

  FIG. 3 is a block diagram showing details of the computer 30 of FIG. In FIG. 3, the camera 10 and the vehicle speed sensor 20 are also illustrated in order to clarify the connection relationship.

  As shown in FIG. 3, the computer 30 includes a viewpoint conversion unit 31, an alignment unit 32, a three-dimensional object detection unit 33, a white turbidity calculation unit 34, a headlight detection unit 35, and a certainty factor calculation unit 36. And a threshold value changing unit 37. Below, each structure is demonstrated.

  The viewpoint conversion unit 31 inputs captured image data of a predetermined area obtained by imaging with the camera 10 and converts the input captured image data into a bird's-eye image data in a bird's-eye view state. The state viewed from a bird's-eye view is a state viewed from the viewpoint of a virtual camera looking down from above, for example, vertically downward. This viewpoint conversion can be executed as described in, for example, Japanese Patent Application Laid-Open No. 2008-219063. The viewpoint conversion of captured image data to bird's-eye view image data is based on the principle that a vertical edge peculiar to a three-dimensional object is converted into a straight line group passing through a specific fixed point by viewpoint conversion to bird's-eye view image data. This is because a planar object and a three-dimensional object can be distinguished if used.

  The alignment unit 32 sequentially inputs the bird's-eye view image data obtained by the viewpoint conversion of the viewpoint conversion unit 31 and aligns the positions of the inputted bird's-eye view image data at different times. 4A and 4B are diagrams for explaining the outline of the processing of the alignment unit 32, where FIG. 4A is a plan view showing the moving state of the host vehicle V1, and FIG. 4B is an image showing the outline of the alignment.

As shown in FIG. 4 (a), the host vehicle V1 of the current time is located in P _1, one unit time before the vehicle V1 is located in the P _{1 '.} Further, there is a parallel running state with the vehicle V1 is located is adjacent vehicle V2 laterally after the vehicle V1, located in P ₂ adjacent vehicle V2 is the current time, one unit time before the adjacent vehicle V2 is P ₂ Suppose it is located at '. Furthermore, it is assumed that the host vehicle V1 has moved a distance d at one time. Note that “one hour before” may be a past time for a predetermined time (for example, one control cycle) from the current time, or may be a past time for an arbitrary time.

In this state, the bird's-eye view image PB _t at the current time is as shown in Figure 4 (b). In the bird's-eye view image PB _t, becomes a rectangular shape for the white line drawn on the road surface, but a relatively accurate is a plan view state, the adjacent vehicle V2 (position P ₂₎ is tilting occurs. Similarly, for the bird's-eye view image PB _{t-1 one} hour before, the white line drawn on the road surface has a rectangular shape and is in a state of being relatively accurately viewed in plan, but the adjacent vehicle V2 (position P _2). ') Will fall down. As described above, the vertical edges of solid objects (including the edges that rise in the three-dimensional space from the road surface in addition to the vertical edges in the strict sense) are straight lines along the collapse direction by the viewpoint conversion processing to bird's-eye view image data. This is because the plane image on the road surface does not include a vertical edge, but such a fall does not occur even when the viewpoint is changed.

The alignment unit 32 performs alignment of the bird's-eye view images PB _t and PB _t−1 as described above on the data. At this time, the alignment unit 32 is offset a bird's-eye view image PB _t-1 before one unit time, to match the position and bird's-eye view image PB _t at the current time. The image on the left side and the center image in FIG. 4B show a state that is offset by the movement distance d ′. This offset amount d ′ is a movement amount on the bird's-eye view image data corresponding to the actual movement distance d of the host vehicle V1 shown in FIG. 4 (a). It is determined based on the time until the time.

  In the present embodiment, the alignment unit 32 aligns the positions of the bird's-eye view images at different times on the bird's-eye view, and obtains the aligned bird's-eye view image. This can be performed with accuracy according to the type of detection target and the required detection accuracy. For example, it may be a strict alignment process that aligns positions based on the same time and the same position, or may be a loose alignment process that grasps the coordinates of each bird's-eye view image.

In addition, after the alignment, the alignment unit 32 calculates the difference between the bird's-eye view images PB _t and PB _t−1 and generates data of the difference image PD _t . Here, in the present embodiment, the alignment unit 32 converts the difference between the pixel values of the bird's-eye view images PB _t and PB _t−1 to an absolute value in order to cope with a change in the illumination environment, and the absolute value is a predetermined value. When the difference value is equal to or greater than the threshold value th, the pixel value of the difference image PD _t is set to “1”, and when the absolute value is less than the predetermined difference threshold value th, the pixel value of the difference image PD _t is set to “0”. Thus, data of the difference image PD _t as shown on the right side of FIG. 4B can be generated. In the present embodiment, the difference threshold th may be changed by a threshold change unit 37 described later. When the difference threshold th is changed by the threshold change unit 37, the difference threshold th is changed by the threshold change unit 37. The pixel value of the difference image PD _t is detected using the difference threshold th.

Then, the three-dimensional object detection unit 33 based on the data of the difference image PD _t shown in FIG. 4 (b), to produce a difference waveform. At this time, the three-dimensional object detection unit 33 also calculates the movement distance of the three-dimensional object in the real space. In detecting the three-dimensional object and calculating the movement distance, the three-dimensional object detection unit 33 first generates a differential waveform.

In generating the difference waveform, the three-dimensional object detection unit 33 sets a detection region (detection frame) in the difference image PD _t . The three-dimensional object detection device 1 of the present example is intended to calculate a movement distance for an adjacent vehicle that may be contacted when the host vehicle V1 changes lanes. For this reason, in this example, as shown in FIG. 2, rectangular detection areas (detection frames) A1 and A2 are set on the rear side of the host vehicle V1. Such detection areas A1, A2 may be set from a relative position with respect to the host vehicle V1, or may be set based on the position of the white line. When setting the position of the white line as a reference, the three-dimensional object detection device 1 may use, for example, an existing white line recognition technique.

  Further, as shown in FIG. 2, the three-dimensional object detection unit 33 recognizes the sides (sides along the traveling direction) of the set detection areas A1 and A2 on the own vehicle V1 side as the ground lines L1 and L2. In general, the ground line means a line in which the three-dimensional object contacts the ground. However, in the present embodiment, the ground line is set as described above, not a line in contact with the ground. Even in this case, from experience, the difference between the ground line according to the present embodiment and the ground line obtained from the position of the original adjacent vehicle V2 is not too large, and there is no problem in practical use.

FIG. 5 is a schematic diagram illustrating how the three-dimensional object detection unit 33 generates a differential waveform. As shown in FIG. 5, the three-dimensional object detection unit 33 calculates a differential waveform from a portion corresponding to the detection areas A <b> 1 and A <b> 2 in the difference image PD _t (right diagram in FIG. 4B) calculated by the alignment unit 32. DW _t is generated. At this time, the three-dimensional object detection unit 33 generates a differential waveform DW _t along the direction in which the three-dimensional object falls by viewpoint conversion. In the example shown in FIG. 5, only the detection area A1 is described for convenience, but the difference waveform DW _t is generated for the detection area A2 in the same procedure.

Specifically, first three-dimensional object detection unit 33 defines a line La on the direction the three-dimensional object collapses on data of the difference image PD _t. Then, the three-dimensional object detection unit 33 counts the number of difference pixels DP indicating a predetermined difference on the line La. In the present embodiment, the difference pixel DP indicating the predetermined difference is expressed by the pixel value of the difference image PD _t as “0” and “1”, and the pixel indicating “1” is counted as the difference pixel DP. .

  The three-dimensional object detection unit 33 counts the number of difference pixels DP, and then obtains an intersection CP between the line La and the ground line L1. Then, the three-dimensional object detection unit 33 associates the intersection CP with the count number, determines the horizontal axis position based on the position of the intersection CP, that is, the position on the vertical axis in the right diagram of FIG. The axis position, that is, the position on the left-right axis in the right diagram of FIG. 5 is determined and plotted as the number of counts at the intersection CP.

Similarly, the three-dimensional object detection unit 33 defines lines Lb, Lc... In the direction in which the three-dimensional object falls, counts the number of difference pixels DP, and determines the horizontal axis position based on the position of each intersection CP. Then, the vertical axis position is determined from the count number (number of difference pixels DP) and plotted. The three-dimensional object detection unit 33 generates the differential waveform DW _t as shown in the right diagram of FIG.

Here, the difference pixel PD on the data of the difference image PD _t is a pixel that has changed in the images at different times, in other words, a location where a three-dimensional object exists. For this reason, the difference waveform DW _t is generated by counting the number of pixels along the direction in which the three-dimensional object collapses and performing frequency distribution at the location where the three-dimensional object exists. In particular, since the number of pixels is counted along the direction in which the three-dimensional object falls, the differential waveform DW _t is generated from the information in the height direction for the three-dimensional object.

As shown in the left diagram of FIG. 5, the line La and the line Lb in the direction in which the three-dimensional object collapses have different distances overlapping the detection area A1. For this reason, if the detection area A1 is filled with the difference pixels DP, the number of difference pixels DP is larger on the line La than on the line Lb. For this reason, when the three-dimensional object detection unit 33 determines the vertical axis position from the count number of the difference pixels DP, the three-dimensional object detection unit 33 is normalized based on the distance at which the lines La and Lb in the direction in which the three-dimensional object falls and the detection area A1 overlap. Turn into. As a specific example, in the left diagram of FIG. 5, there are six difference pixels DP on the line La, and there are five difference pixels DP on the line Lb. For this reason, in determining the vertical axis position from the count number in FIG. 5, the three-dimensional object detection unit 33 normalizes the count number by dividing it by the overlap distance. Thus, as shown in the difference waveform DW _t, the line La on the direction the three-dimensional object collapses, the value of the differential waveform DW _t corresponding to Lb is substantially the same.

After the generation of the difference waveform DW _t , the three-dimensional object detection unit 33 detects the adjacent vehicle existing in the adjacent lane based on the generated difference waveform DW _t . Here, FIG. 6 is a diagram for explaining a method of detecting a three-dimensional object by the three-dimensional object detection unit 33, and illustrates an example of the difference waveform DW _t and a threshold value α for detecting the three-dimensional object. As illustrated in FIG. 6, the three-dimensional object detection unit 33 determines whether or not the peak of the generated differential waveform DW _t is _equal to or greater than a predetermined threshold value α corresponding to the peak position of the differential waveform DW _t. Then, it is determined whether or not a three-dimensional object exists in the detection areas A1 and A2. Then, when the peak of the difference waveform DW _t is less than the predetermined threshold value α, the three-dimensional object detection unit 33 determines that there is no three-dimensional object in the detection areas A1 and A2, while the peak of the difference waveform DW _t Is greater than or equal to a predetermined threshold value α, it is determined that a three-dimensional object exists in the detection areas A1 and A2.

Moreover, three-dimensional object detection unit 33, in contrast with the differential waveform DW _t-1 of the previous differential waveform DW _t and a time instant at the current time, and calculates the moving speed of the three-dimensional object. That is, the three-dimensional object detection unit 33 calculates the moving speed of the three-dimensional object from the time change of the differential waveforms DW _t and DW _t−1 .

Specifically, as shown in FIG. 7, the three-dimensional object detection unit 33 divides the differential waveform DW _t into a plurality of small areas DW _{t1 to} DW _tn (n is an arbitrary integer equal to or greater than 2). FIG. 7 is a diagram illustrating the small areas DW _{t1 to} DW _tn divided by the three-dimensional object detection unit 33. The small areas DW _{t1 to} DW _tn are divided so as to overlap each other, for example, as shown in FIG. For example, the small area DW _t1 and the small area DW _t2 overlap, and the small area DW _t2 and the small area DW _t3 overlap.

Next, the three-dimensional object detection unit 33 obtains an offset amount (amount of movement of the differential waveform in the horizontal axis direction (vertical direction in FIG. 7)) for each of the small regions DW _{t1 to} DW _tn . Here, the offset amount is determined from the difference between the differential waveform DW _t in the difference waveform DW _t-1 and the current time before one unit time (distance in the horizontal axis direction). At this time, three-dimensional object detection unit 33, for each small area _DW t1 _{~DW tn,} when moving the differential waveform _{DW t1} before one unit time in the horizontal axis direction, the differential waveform DW _t at the current time The position where the error is minimized (the position in the horizontal axis direction) is determined, and the amount of movement in the horizontal axis between the original position of the differential waveform DW _t-1 and the position where the error is minimized is obtained as an offset amount. Then, the three-dimensional object detection unit 33 counts the offset amount obtained for each of the small areas DW _{t1 to} DW _tn and forms a histogram.

FIG. 8 is a diagram illustrating an example of a histogram obtained by the three-dimensional object detection unit 33. As shown in FIG. 8, there is some variation in the offset amount, which is the amount of movement that minimizes the error between each of the small areas DW _{t1 to} DW _tn and the differential waveform DW _t-1 one time before. For this reason, the three-dimensional object detection unit 33 forms a histogram of offset amounts including variations, and calculates a movement distance from the histogram. At this time, the three-dimensional object detection unit 33 calculates the moving distance of the three-dimensional object (adjacent vehicle V2) from the maximum value of the histogram. That is, in the example illustrated in FIG. 8, the three-dimensional object detection unit 33 calculates the offset amount indicating the maximum value of the histogram as the movement distance τ ^* . As described above, in this embodiment, even if the offset amount varies, it is possible to calculate a more accurate movement distance from the maximum value. The moving distance τ ^* is a relative moving distance of the three-dimensional object (adjacent vehicle V2) with respect to the own vehicle. For this reason, when calculating the absolute movement distance, the three-dimensional object detection unit 33 calculates the absolute movement distance based on the obtained movement distance τ ^* and the signal from the vehicle speed sensor 20.

As described above, in the present embodiment, by calculating the moving distance of the three-dimensional object (adjacent vehicle V2) from the offset amount of the differential waveform DW _t when the error of the differential waveform DW _t generated at different times is minimized. Therefore, the movement distance is calculated from the offset amount of the one-dimensional information called the waveform, and the calculation cost can be suppressed in calculating the movement distance. In addition, by dividing the differential waveform DW _t generated at different times into a plurality of small regions DW _{t1 to} DW _tn , a plurality of waveforms representing the respective locations of the three-dimensional object can be obtained. Since the offset amount can be obtained for each of the positions, and the movement distance can be obtained from a plurality of offset amounts, the calculation accuracy of the movement distance can be improved. Further, in the present embodiment, by calculating the moving distance of the three-dimensional object from the time change of the differential waveform DW _t including the information in the height direction, compared with a case where attention is paid only to one point of movement, Since the detection location before the time change and the detection location after the time change are specified including information in the height direction, it is likely to be the same location in the three-dimensional object, and the movement distance is calculated from the time change of the same location, and the movement Distance calculation accuracy can be improved.

Incidentally, the three-dimensional object detection unit 33 Upon histogram is weighted for each of a plurality of small areas _DW t1 _{~DW tn,} the offset amount determined for each small area _DW t1 _{~DW tn} histogram of counts in response to the weight May be. FIG. 9 is a diagram illustrating weighting by the three-dimensional object detection unit 33.

As shown in FIG. 9, the small area DW _m (m is an integer of 1 to n−1) is flat. That is, in the small area DW _m , the difference between the maximum value and the minimum value of the number of pixels indicating a predetermined difference is small. Three-dimensional object detection unit 33 to reduce the weight for such small area DW _m. This is because the flat small area DW _m has no characteristics and is likely to have a large error in calculating the offset amount.

On the other hand, the small area DW _{m + k} (k is an integer equal to or less than nm) is rich in undulations. That is, in the small area DW _m , the difference between the maximum value and the minimum value of the number of pixels indicating a predetermined difference is large. Three-dimensional object detection unit 33 increases the weight for such small area DW _m. This is because the small region DW _{m + k} rich in undulations is characteristic and there is a high possibility that the offset amount can be accurately calculated. By weighting in this way, the calculation accuracy of the movement distance can be improved.

Although dividing the differential waveform DW _t into a plurality of small areas DW _t1 ~DW _tn in the above embodiment in order to improve the calculation accuracy of the moving distance, if the calculation accuracy of the moving distance is not less required small regions DW _t1 It is not necessary to divide into ~ DW _tn . In this case, the three-dimensional object detection unit 33 calculates the movement distance from the offset amount of the differential waveform DW _t when the error between the differential waveform DW _t and the differential waveform DW _t−1 is minimized. That is, the method for obtaining the offset amount of the difference waveform DW _t in the difference waveform DW _t-1 and the current time before one unit time is not limited to the above disclosure.

  In the present embodiment, the three-dimensional object detection unit 33 obtains the moving speed of the host vehicle V1 (camera 10), and obtains the offset amount for the stationary object from the obtained moving speed. After obtaining the offset amount of the stationary object, the three-dimensional object detection unit 33 calculates the moving distance of the three-dimensional object after ignoring the offset amount corresponding to the stationary object among the maximum values of the histogram.

  FIG. 10 is a diagram illustrating another example of a histogram obtained by the three-dimensional object detection unit 33. When a stationary object is present in addition to a three-dimensional object within the angle of view of the camera 10, two maximum values τ1 and τ2 appear in the obtained histogram. In this case, one of the two maximum values τ1, τ2 is the offset amount of the stationary object. For this reason, the three-dimensional object detection unit 33 calculates the offset amount for the stationary object from the moving speed, ignores the maximum value corresponding to the offset amount, and calculates the moving distance of the three-dimensional object using the remaining maximum value. To do. Thereby, the situation where the calculation accuracy of the moving distance of a solid object falls by a stationary object can be prevented.

  Even if the offset amount corresponding to the stationary object is ignored, if there are a plurality of maximum values, it is assumed that there are a plurality of three-dimensional objects within the angle of view of the camera 10. However, it is extremely rare that a plurality of three-dimensional objects exist in the detection areas A1 and A2. For this reason, the three-dimensional object detection unit 33 stops calculating the movement distance. Thereby, in the present embodiment, it is possible to prevent a situation in which an erroneous movement distance having a plurality of maximum values is calculated.

  The white turbidity calculating unit 34 calculates the degree of white turbidity (a white thin film is formed on the lens surface) as a result of adhesion of foreign matter such as scale to the lens of the camera 10 as the white turbidity of the lens. To do. The method for calculating the white turbidity by the white turbidity calculating unit 34 is not particularly limited, and for example, the white turbidity of the lens can be calculated using a method described below.

  For example, when the brightness difference between adjacent pixels is equal to or greater than a predetermined value, the white turbidity calculation unit 34 determines that an object edge (outline) exists between these adjacent pixels, and such An edge of a subject is detected from pixels having a luminance difference. At this time, the white turbidity calculating unit 34 detects the intensity of the edge of the subject with a higher value as the luminance difference between the pixels is larger, and determines that the captured image is clearer as the detected edge intensity is higher. Then, the sharpness of the image is calculated with a high value. Then, the white turbidity calculating unit 34 can determine that the lens is not clouded as the captured image is clear and the sharpness of the image is high, and can calculate the white turbidity of the lens with a small value.

  Here, FIG. 11 is a diagram for explaining the relationship between the white turbidity of the lens and the sharpness of the image. In FIG. 11, the vertical axis indicates the luminance, and the horizontal axis indicates the pixel position. FIG. 11 illustrates a scene in which a high-brightness subject such as a streetlight or a headlight of an adjacent vehicle is imaged. As shown in FIG. 11, when the lens is clouded, a part of light incident on the lens from the subject is blocked or irregularly reflected by foreign matters such as dirt adhering to the lens. Compared with the case where is not cloudy, the gradient of the luminance peak becomes gentler, and the luminance difference between the image areas becomes smaller. For this reason, when the lens is clouded, the edge intensity detected from the captured image becomes small, and the sharpness of the image is calculated with a low value. On the other hand, when the lens is not clouded, as shown in FIG. 11, the gradient of the luminance peak becomes steep, and the luminance difference between the image areas increases. For this reason, when the lens is not clouded, the edge intensity detected by the cloudiness calculation unit 34 is increased, thereby calculating the image sharpness with a high value. Therefore, the white turbidity calculating unit 34 calculates the sharpness of the image based on the edge strength detected from the captured image. As a result, the lower the sharpness of the image, the higher the white turbidity of the lens. In addition, the higher the image sharpness, the lower the turbidity of the lens.

  Further, the white turbidity calculating unit 34 can also calculate the white turbidity of the lens by calculating the sharpness of the image based on the luminance gradient around the high luminance region having the luminance equal to or higher than a predetermined value. That is, as shown in FIG. 11, when the lens is not clouded, the brightness gradient becomes steep, while when the lens is clouded, the brightness gradient becomes gentle. Therefore, the white turbidity calculating unit 34 detects a gradient of luminance from the outer edge of the high luminance region corresponding to the light source such as the streetlight or the headlight to the outside in the image region including the image of the light source such as the streetlight or the headlight. When a steep brightness gradient is detected, it is judged that the image has high sharpness and the lens is not cloudy, and the lens cloudiness is calculated with a low value, while a gentle brightness gradient is detected. In this case, it is determined that the sharpness of the image is low and the lens is clouded, and the cloudiness of the lens is calculated as a high value.

  In addition, the white turbidity calculation unit 34 can also detect the white turbidity of the lens based on the frequency component of the image. That is, the frequency component of the subject is detected from the image signal, and the high frequency component is removed by applying a low pass filter process to the detected frequency component. Then, by comparing the frequency component after removing the high frequency component with the frequency component before removing the high frequency component, the high frequency component of the subject is detected, and the detected high frequency component is calculated as the sharpness of the image. Can do. For example, since the high-frequency component of the subject is obtained from a high-contrast region, the image becomes clearer (high contrast) as the high-frequency component increases, and in this case, the white turbidity of the lens can be calculated with a low value.

  Furthermore, the white turbidity calculation unit 34 can also calculate the white turbidity of the lens by calculating the sharpness of the image based on the size of the detection reference value used for detecting the edge from the captured image. . Here, FIG. 12 is a diagram for explaining a method of calculating the sharpness of an image based on the detection reference value. For example, in the example illustrated in FIG. 12, the turbidity calculation unit 34 first detects an edge whose luminance difference between adjacent pixels is equal to or greater than a preset detection reference value ts, and detects an edge that is greater than or equal to a predetermined amount. It is determined whether it has been completed. Then, when an edge exceeding a predetermined amount cannot be detected, the turbidity calculation unit 34 changes the detection reference value ts to a small value, and detects an edge where the luminance difference between adjacent pixels is equal to or more than the changed detection reference value. Then, it is determined whether or not a predetermined amount or more of edges have been detected. In this way, the turbidity calculating unit 34 repeatedly detects edges while changing the detection reference value to a small value, and can detect a detection reference value ts ′ (detecting an edge greater than a predetermined amount). The largest detection reference value ts ′) among the detection reference values is specified. Here, when the same subject is imaged, as shown in FIG. 12, the lens is clouded and the outline of the subject becomes blurred and the detected edge strength of the subject becomes smaller as the image becomes unclear. Therefore, as shown in FIG. 12, when the lens is clouded and the image is unclear, it is impossible to detect an edge of a predetermined amount or more unless the detection reference value is changed to a smaller value. Accordingly, the white turbidity calculating unit 34 determines that the lens becomes cloudy and the image is unclear as the detection reference value that has detected an edge of a predetermined amount or more is small, and calculates the white turbidity of the lens as a high value. Can do.

  Returning to FIG. 3, the headlight detection unit 35 detects a light source that is a candidate for a headlight of the adjacent vehicle V <b> 2 that exists behind the host vehicle V <b> 1 based on the captured image captured by the camera 10. Specifically, the headlight detection unit 35 selects an image area whose brightness difference from the surrounding area is equal to or greater than a predetermined value and whose size is equal to or greater than a predetermined area, corresponding to the headlight of the adjacent vehicle V2. By detecting as a region, a light source that is a candidate for a headlight of the adjacent vehicle V2 is detected.

  The certainty factor calculation unit 36 calculates the certainty factor that the light source detected by the headlight detection unit 35 is the headlight of the adjacent vehicle V2. Here, compared to the headlight of the adjacent vehicle V2 that travels in the adjacent lane adjacent to the traveling lane of the host vehicle V1, the headlight of the adjacent vehicle that travels in the adjacent adjacent lane two adjacent to the traveling vehicle of the host vehicle V1. A light source such as a streetlight installed outside the road is detected at a position away from the vehicle V1. For this reason, the certainty factor calculation unit 36 increases the certainty factor that the light source detected by the headlight detection unit 35 is the headlight of the adjacent vehicle V2 as the distance in the vehicle width direction from the host vehicle V1 to the light source is shorter. Can be calculated. Further, since the larger the light source size detected by the headlight detection unit 35 is, the more the light source can be determined to be located closer to the host vehicle V1, the certainty factor calculation unit 36 is detected by the headlight detection unit 35. The greater the size of the light source, the higher the certainty that the light source detected by the headlight detector 35 is the headlight of the adjacent vehicle V2. Furthermore, while the light source such as a streetlight is a stationary object and does not move, the adjacent vehicle V2 moves. The moving speed of the headlight of the adjacent vehicle V2 is higher than the moving speed of the light source of the stationary object such as a streetlight. Will also be faster. Further, since the headlight of the adjacent vehicle is detected at a position farther from the own vehicle V1 than the headlight of the adjacent vehicle V2, the headlight of the adjacent vehicle V2 is detected rather than the moving speed of the headlight of the adjacent vehicle. There is a tendency that the moving speed of is faster. Therefore, the certainty factor calculation unit 36 calculates the movement speed of the light source based on the change in the position of the light source detected by the headlight detection unit 35, and the higher the movement speed of the light source, the more the light source moves to the adjacent vehicle V2. The certainty that it is a headlight can be calculated high.

  The threshold value changing unit 37 changes the difference threshold value th based on the certainty factor of the headlight of the adjacent vehicle V2 calculated by the certainty factor calculating unit 36. Here, FIG. 13 is an example of a map showing the relationship between the reliability of the headlight of the adjacent vehicle V2 and the difference threshold th. As shown in FIG. 13, the threshold value changing unit 37 changes the difference threshold value th to a lower value as the certainty level that the light source detected by the headlight detecting unit 35 is the headlight of the adjacent vehicle V2 is higher. . In other words, in the present embodiment, when the certainty that the detected light source is the headlight of the adjacent vehicle V2 is high, it is determined that the adjacent vehicle V2 is likely to exist, and the difference threshold th is set as the difference threshold th. By changing to a low value and facilitating detection of a three-dimensional object (adjacent vehicle V2) existing in the adjacent lane, it is possible to appropriately detect the adjacent vehicle V2 even when the lens is clouded. On the other hand, when the certainty that the detected light source is the headlight of the adjacent vehicle V2 is low, it is determined that there is a high possibility that the adjacent vehicle V2 does not exist, and the difference threshold th is changed to a high value. By suppressing the detection of a three-dimensional object (adjacent vehicle V2) present in the lane, when the lens is clouded, a foreign object adhering to the lens or a three-dimensional object other than the adjacent vehicle V2 is mistaken as the adjacent vehicle V2. It is possible to effectively prevent the detection.

  Further, the calculation unit 30 stores a plurality of maps indicating the relationship between the difference threshold th and the certainty of the headlights of the adjacent vehicle V2 as shown in FIG. 13 in a memory (not shown). A map corresponding to the turbidity of the lens is selected from the plurality of maps. For example, in FIG. 13, an example of a map of the difference threshold th selected when the white turbidity of the lens is high and the certainty of the headlight of the adjacent vehicle V2 is shown by a wavy line, and the white turbidity of the lens is low An example of a map of the difference threshold th selected in step S3 and the certainty of the headlight of the adjacent vehicle V2 is shown by a solid line. As shown in FIG. 13, the threshold change unit 37 sets the difference threshold th to a lower value as the turbidity of the lens is higher, even if the certainty that the lens is a headlight of the adjacent vehicle V2 is low. The map indicating the relationship between the difference threshold th and the certainty of the headlight of the adjacent vehicle V2 is selected.

That is, in the example shown in FIG. 13, when the white turbidity of the lens is high, a map displayed with a broken line is selected, and when the white turbidity of the lens is low, a map displayed with a solid line is selected. Therefore, when the certainty of the headlight of the adjacent vehicle V2 is, for example, _Sc1 shown in FIG. 13, when the white turbidity of the lens is low, the difference threshold is set to _a predetermined value th _a and the white turbidity of the lens is high. In this case, the difference threshold th _b is set lower than the difference threshold th _b when the white turbidity of the lens is low. Thus, in this embodiment, the higher the white turbidity of the lens, the higher the certainty that the detected light source is the headline of the adjacent vehicle V2, and the certainty of the calculated headline of the adjacent vehicle V2. Even if the degrees are the same value, a map indicating the relationship between the difference threshold th and the certainty of the headlight of the adjacent vehicle V2 is selected so that the difference threshold th is set to a low value.

  Here, when the lens is clouded, a part of the light beam received from the headlight of the adjacent vehicle V2 is blocked or diffusely reflected by a foreign matter such as scale adhering to the lens. The light of the headlight of the adjacent vehicle V2 received at is attenuated. In particular, since the amount of light decreases as the distance from the center of the headlight decreases, when the lens is clouded, the camera 10 can sufficiently receive the light at the periphery of the headlight due to foreign matters such as water scales adhering to the lens. In other words, the headlight light imaged may be smaller than when the lens is not clouded. In such a case, even if the light source detected by the headlight detection unit 35 is actually the headlight of the adjacent vehicle V2, if the lens is clouded, the light source detected by the headlight detection unit 35 Is less certain that it is a headlight of the adjacent vehicle V2. Therefore, when the lens is clouded, the threshold value changing unit 37 uses the headlight of the adjacent vehicle V2 as the light source detected by the headlight detection unit 35 even when the certainty of the headlight of the adjacent vehicle V2 is low. It is determined that the certainty factor is high, and the map indicating the relationship between the difference threshold value th and the certainty factor of the headlight of the adjacent vehicle V2 is changed so that the difference threshold value th is set low. Thereby, even when the white turbidity of the lens is high and it is difficult to detect the headlight of the adjacent vehicle V2, the light source detected by the headlight detection unit 35 is appropriately determined as the headlight of the adjacent vehicle V2. As a result, it is possible to appropriately detect the adjacent vehicle V2.

  When the opacity of the lens is high, light sources such as headlights and street lights of the adjacent vehicle V2 are harder to detect than the headlights of the adjacent vehicle V2. Even when the map showing the relationship between the difference threshold th and the certainty of the headlight of the adjacent vehicle V2 is changed so that the difference threshold th becomes a low value, the headlight, streetlight, etc. of the adjacent vehicle V2 are changed to the adjacent vehicle V2. Can be prevented from being erroneously detected.

Next, the adjacent vehicle detection process according to the present embodiment will be described. FIG. 14 is a flowchart illustrating the adjacent vehicle detection process of the first embodiment. As shown in FIG. 14, first, the computer 30 acquires captured image data from the camera 10 (step S <b> 101), and the viewpoint conversion unit 31 acquires the bird's-eye view image PB based on the acquired captured image data. Data of _t is generated (step S102).

Next, the alignment unit 32 aligns the data of the bird's-eye view image PB _{t and} the data of the bird's-eye view image PB _{t-1 one} hour before, and generates data of the difference image PD _t (step S103). . Specifically, the alignment unit 32 converts the difference between the pixel values of the bird's-eye view images PB _t and PB _t−1 to an absolute value, and when the absolute value is equal to or greater than a predetermined difference threshold th, the difference image PD _t Is set to “1”, and when the absolute value is less than the predetermined difference threshold th, the pixel value of the difference image PD _t is set to “0”. Note that the difference threshold th for calculating the pixel value of the difference image PD _t may be changed in a threshold change process described later. When the difference threshold th is changed, the changed difference threshold th is This is used in step S103. Then, three-dimensional object detection unit 33, from the data of the difference image PD _t, pixel value by counting the number of difference pixel DP "1", to generate a difference waveform DW _t (step S104).

Then, the three-dimensional object detection unit 33 determines whether or not the peak of the differential waveform DW _t is greater than or _{equal to} a predetermined threshold value α (step S105). When the peak of the difference waveform DW _t is not equal to or greater than the threshold value α, that is, when there is almost no difference, it is considered that there is no three-dimensional object in the captured image. For this reason, when it is determined that the peak of the differential waveform DW _t is not equal to or greater than the threshold value α (step S105 = No), the three-dimensional object detection unit 33 determines that there is no three-dimensional object and there is no adjacent vehicle V2 ( Step S114). And it returns to step S101 and repeats the process shown in FIG.

On the other hand, if it is determined that the peak of the difference waveform DW _t is equal to or greater than the threshold value α (step S105 = Yes), the three-dimensional object detection unit 33 determines that a three-dimensional object is present in the adjacent lane, and proceeds to step S106. The three-dimensional object detection unit 33 divides the differential waveform DW _t into a plurality of small areas DW _{t1 to} DW _tn . Next, the three-dimensional object detection unit 33 performs weighting for each of the small areas DW _{t1 to} DW _tn (Step S107), calculates an offset amount for each of the small areas DW _{t1 to} DW _tn (Step S108), and adds the weight. A histogram is generated (step S109).

  Then, the three-dimensional object detection unit 33 calculates a relative movement distance that is a movement distance of the three-dimensional object with respect to the host vehicle V1 based on the histogram (step S110). Next, the three-dimensional object detection unit 33 calculates the absolute movement speed of the three-dimensional object from the relative movement distance (step S111). At this time, the three-dimensional object detection unit 33 calculates the relative movement speed by differentiating the relative movement distance with respect to time, and calculates the absolute movement speed by adding the own vehicle speed detected by the vehicle speed sensor 20.

  Thereafter, the three-dimensional object detection unit 33 determines whether or not the absolute movement speed of the three-dimensional object is 10 km / h or more and the relative movement speed of the three-dimensional object with respect to the host vehicle V1 is +60 km / h or less (step S112). . When both are satisfied (step S112 = Yes), the three-dimensional object detection unit 33 determines that the detected three-dimensional object is the adjacent vehicle V2 existing in the adjacent lane and the adjacent vehicle V2 exists in the adjacent lane (step S113). ). Then, the process shown in FIG. 14 ends. On the other hand, when either one is not satisfied (step S112 = No), the three-dimensional object detection unit 33 determines that the adjacent vehicle V2 does not exist in the adjacent lane (step S114). And it returns to step S101 and repeats the process shown in FIG.

  In the present embodiment, the left and right rear sides of the host vehicle V1 are set as detection areas A1 and A2, and emphasis is placed on whether or not there is a possibility of contact when the host vehicle V1 changes lanes. For this reason, the process of step S112 is performed. That is, assuming that the system according to this embodiment is operated on a highway, when the speed of the adjacent vehicle V2 is less than 10 km / h, even when the adjacent vehicle V2 exists, when changing the lane, Since it is located far behind the host vehicle V1, there is little problem. Similarly, when the relative movement speed of the adjacent vehicle V2 with respect to the own vehicle V1 exceeds +60 km / h (that is, when the adjacent vehicle V2 moves at a speed higher than 60 km / h than the speed of the own vehicle V1), the lane When changing, it is less likely to cause a problem because the vehicle is moving in front of the host vehicle V1. For this reason, in step S112, it can be said that the adjacent vehicle V2 which becomes a problem at the time of lane change is judged.

  Moreover, the following effects are obtained by determining whether the absolute moving speed of the adjacent vehicle V2 is 10 km / h or more and the relative moving speed of the adjacent vehicle V2 with respect to the host vehicle V1 is +60 km / h or less in step S112. . For example, depending on the mounting error of the camera 10, the absolute moving speed of the stationary object may be detected to be several km / h. Therefore, by determining whether the speed is 10 km / h or more, it is possible to reduce the possibility of determining that the stationary object is the adjacent vehicle V2. Further, depending on the noise, the relative speed of the adjacent vehicle V2 with respect to the host vehicle V1 may be detected as a speed exceeding +60 km / h. Therefore, the possibility of erroneous detection due to noise can be reduced by determining whether the relative speed is +60 km / h or less.

  Furthermore, instead of the process of step S112, it may be determined that the absolute movement speed of the adjacent vehicle V2 is not negative or 0 km / h. Further, in the present embodiment, since emphasis is placed on whether or not there is a possibility of contact when the host vehicle V1 changes lanes, when the adjacent vehicle V2 is detected in step S113, the driver of the host vehicle is notified. A warning sound may be emitted or a display corresponding to a warning may be performed by a predetermined display device.

  Next, the threshold value changing process according to the first embodiment will be described. FIG. 15 is a flowchart showing threshold change processing according to the first embodiment. The threshold change process described below is performed in parallel with the adjacent vehicle detection process shown in FIG. 14, and the difference threshold th set by this threshold change process is the difference threshold in the adjacent vehicle detection process shown in FIG. It will be applied as th.

  As shown in FIG. 15, first, in step S201, the headlight detection unit 35 detects a light source that is a candidate for a headlight of the adjacent vehicle V2. Specifically, the headlight detection unit 35 corresponds to the light source of the headlight of the adjacent vehicle V2 for an image area in which the brightness difference from the surroundings is equal to or greater than a predetermined value and the size is greater than or equal to the predetermined value. By detecting as a candidate area, a light source that is a candidate for a headlight of the adjacent vehicle V2 is detected.

  In step S202, the headlight detection unit 35 determines whether a light source that is a candidate for a headlight of the adjacent vehicle V2 is detected based on the detection result in step S201. If a light source that is a candidate for the headlight of the adjacent vehicle V2 is detected, the process proceeds to step S203. On the other hand, if a light source that is a candidate for the headlight of the adjacent vehicle V2 is not detected, the process returns to step S201. The detection of the light source that becomes the headlight of the adjacent vehicle V2 is repeatedly performed.

  In step S203, the certainty factor calculation unit 36 calculates the certainty factor that the light source detected in step S201 is the headlight of the adjacent vehicle V2. Specifically, the certainty factor calculation unit 36 determines the vehicle width direction from the detected light source to the host vehicle V1 as the moving speed of the light source detected in step S201 is faster or the detected light source is larger. The shorter the distance at, the higher the certainty that the detected light source is the headlight of the adjacent vehicle V2.

  In step S204, the white turbidity calculation unit 34 calculates the white turbidity indicating the degree of white turbidity of the lens. For example, as shown in FIG. 11, the white turbidity calculation unit 34 can calculate the white turbidity of the lens by calculating the sharpness of the image based on the intensity of the edge detected from the captured image. In this case, the white turbidity calculating unit 34 determines that the image is clearer as the edge intensity detected from the captured image is stronger (the luminance difference between pixels is larger), and the lens is not clouded. The cloudiness of can be calculated small.

  In step S205, the threshold changing unit 37 selects a map indicating the relationship between the certainty of the headlight of the adjacent vehicle V2 and the difference threshold th based on the white turbidity of the lens calculated in step S204. Specifically, as shown in FIG. 13, the threshold value changing unit 37 shows a map in which the difference threshold th is set lower as the turbidity of the lens is higher, even when the certainty of the headlight of the adjacent vehicle V2 is lower. select.

  In step S206, the threshold value changing unit 37 changes the difference threshold value th based on the certainty of the headlight of the adjacent vehicle V2 calculated in step S203 and the map selected in step S205. Specifically, the threshold change unit 37 changes the difference threshold th to a lower value as the certainty that the detected light source is the headlight of the adjacent vehicle V2 is higher according to the map selected in step S205. . Thereby, even when the lens is clouded, when the certainty factor that the light source detected in step S201 is the headlight of the adjacent vehicle V2 is high and it can be determined that the adjacent vehicle V2 exists in the adjacent lane, The difference threshold th for detecting the three-dimensional object (adjacent vehicle V2) is changed to a low value, and as a result, the adjacent vehicle V2 existing in the adjacent lane can be appropriately detected. Then, after the difference threshold th is changed in step S206, the process returns to step S201, and the threshold value changing process described above is repeated.

  As described above, in the first embodiment, the headlight of the adjacent vehicle V2 is detected, and the certainty that the detected light source is the headlight of the adjacent vehicle V2 is calculated. And the difference threshold th is changed to a low value, so that the certainty degree of the headlight of the adjacent vehicle V2 is high. Thereby, when the certainty factor that the detected light source is the headlight of the adjacent vehicle V2 is high and it can be determined that the adjacent vehicle V2 exists in the adjacent lane, the difference threshold th for detecting the three-dimensional object is low. Since the three-dimensional object (adjacent vehicle V2) existing in the adjacent lane is easily detected, it is present in the adjacent lane even if foreign matter such as scale adheres to the lens and the lens is clouded. It is possible to appropriately detect the adjacent vehicle V2.

  In the present embodiment, the degree of white turbidity of the lens is calculated as the white turbidity of the lens, and the higher the white turbidity of the lens, the higher the certainty that the detected light source is the headlight of the adjacent vehicle V2. Thus, the map between the certainty of the headlight of the adjacent vehicle V2 and the difference threshold th is changed. Thereby, for example, even if the detected light source is actually the headlight of the adjacent vehicle V2, a part of the light beam from the headlight of the adjacent vehicle V2 is blocked by foreign matter such as water scale adhering to the lens. Even if the certainty that the detected light source is the headlight of the adjacent vehicle V2 becomes low because the image of the light of the headlight of the adjacent vehicle V2 received by the camera 10 becomes small, it is detected. Therefore, it is possible to appropriately determine that the light source is the headlight of the adjacent vehicle V2, and thus it is possible to appropriately detect the adjacent vehicle V2.

<< Second Embodiment >>
Next, the three-dimensional object detection device 1a according to the second embodiment will be described. As shown in FIG. 16, the three-dimensional object detection device 1 a according to the second embodiment includes a computer 30 a instead of the computer 30 of the first embodiment, except that it operates as described below. This is the same as in the first embodiment. Here, FIG. 16 is a block diagram showing details of the computer 30a according to the second embodiment.

  As shown in FIG. 16, the three-dimensional object detection apparatus 1a according to the second embodiment includes a camera 10 and a computer 30a. The computer 30a includes a viewpoint conversion unit 31, a luminance difference calculation unit 38, and an edge line detection unit. 39, a three-dimensional object detection unit 33a, a white turbidity calculation unit 34, a headlight detection unit 35, a certainty factor calculation unit 36, and a threshold value change unit 37a. Below, each structure of the solid-object detection apparatus 1a which concerns on 2nd Embodiment is demonstrated. Note that the viewpoint conversion unit 31, the white turbidity calculation unit 34, the headlight detection unit 35, and the certainty calculation unit 36 have the same configuration as that of the first embodiment, and thus description thereof is omitted.

  FIG. 17 is a diagram illustrating an imaging range and the like of the camera 10 of FIG. 16, FIG. 17A is a plan view, and FIG. 17B is a perspective view in real space on the rear side from the host vehicle V1. Show. As shown in FIG. 17A, the camera 10 has a predetermined angle of view a, and images the rear side from the host vehicle V1 included in the predetermined angle of view a. Similarly to the case shown in FIG. 2, the angle of view a of the camera 10 is set so that the imaging range of the camera 10 includes the adjacent lane in addition to the lane in which the host vehicle V1 travels.

Detection area A1, A2 of the present embodiment is in a plan view (a state of being bird's view) a trapezoidal shape, location of the detection areas A1, A2, size and shape, based on the distance d ₁ to d ₄ determines Is done. The detection areas A1 and A2 in the example shown in the figure are not limited to a trapezoidal shape, and may be other shapes such as a rectangle when viewed from a bird's eye view as shown in FIG.

  Here, the distance d1 is a distance from the host vehicle V1 to the ground lines L1 and L2. The ground lines L1 and L2 mean lines on which a three-dimensional object existing in the lane adjacent to the lane in which the host vehicle V1 travels contacts the ground. In the present embodiment, the object is to detect adjacent vehicles V2 and the like (including two-wheeled vehicles) traveling in the left and right lanes adjacent to the lane of the host vehicle V1 on the rear side of the host vehicle V1. For this reason, a distance d1 which is a position to be the ground lines L1 and L2 of the adjacent vehicle V2 from a distance d11 from the own vehicle V1 to the white line W and a distance d12 from the white line W to a position where the adjacent vehicle V2 is predicted to travel It can be determined substantially fixedly.

  Further, the distance d1 is not limited to being fixedly determined, and may be variable. In this case, the computer 30a recognizes the position of the white line W with respect to the host vehicle V1 by a technique such as white line recognition, and determines the distance d11 based on the recognized position of the white line W. Thereby, the distance d1 is variably set using the determined distance d11. In the following embodiment, since the position where the adjacent vehicle V2 travels (distance d12 from the white line W) and the position where the host vehicle V1 travels (distance d11 from the white line W) are roughly determined, the distance d1 is It shall be fixedly determined.

  The distance d2 is a distance extending in the vehicle traveling direction from the rear end portion of the host vehicle V1. The distance d2 is determined so that the detection areas A1 and A2 are at least within the angle of view a of the camera 10. In particular, in the present embodiment, the distance d2 is set so as to be in contact with the range divided into the angle of view a. The distance d3 is a distance indicating the length of the detection areas A1, A2 in the vehicle traveling direction. This distance d3 is determined based on the size of the three-dimensional object to be detected. In the present embodiment, since the detection target is the adjacent vehicle V2 or the like, the distance d3 is set to a length including the adjacent vehicle V2.

  As shown in FIG. 17B, the distance d4 is a distance indicating a height that is set to include a tire such as the adjacent vehicle V2 in the real space. The distance d4 is a length shown in FIG. 17A in the bird's-eye view image. Note that the distance d4 may be a length that does not include a lane that is further adjacent to the left and right lanes in the bird's-eye view image (that is, the adjacent lane that is adjacent to two lanes). If the lane adjacent to the two lanes is included from the lane of the own vehicle V1, there is an adjacent vehicle V2 in the adjacent lane on the left and right of the own lane that is the lane in which the own vehicle V1 is traveling. This is because it becomes impossible to distinguish whether there is an adjacent vehicle on the lane.

  As described above, the distances d1 to d4 are determined, and thereby the positions, sizes, and shapes of the detection areas A1 and A2 are determined. More specifically, the position of the upper side b1 of the detection areas A1 and A2 forming a trapezoid is determined by the distance d1. The starting point position C1 of the upper side b1 is determined by the distance d2. The end point position C2 of the upper side b1 is determined by the distance d3. The side b2 of the detection areas A1 and A2 having a trapezoidal shape is determined by a straight line L3 extending from the camera 10 toward the starting point position C1. Similarly, a side b3 of trapezoidal detection areas A1 and A2 is determined by a straight line L4 extending from the camera 10 toward the end position C2. The position of the lower side b4 of the detection areas A1 and A2 having a trapezoidal shape is determined by the distance d4. Thus, the area surrounded by the sides b1 to b4 is set as the detection areas A1 and A2. As shown in FIG. 17B, the detection areas A1 and A2 are true squares (rectangles) in real space on the rear side from the host vehicle V1.

  Returning to FIG. 16, the viewpoint conversion unit 31 inputs captured image data of a predetermined region obtained by imaging with the camera 10. The viewpoint conversion unit 31 performs viewpoint conversion processing on the input captured image data to the bird's-eye image data in a bird's-eye view state. The bird's-eye view is a state seen from the viewpoint of a virtual camera looking down from above, for example, vertically downward (or slightly obliquely downward). This viewpoint conversion process can be realized by a technique described in, for example, Japanese Patent Application Laid-Open No. 2008-219063.

  The luminance difference calculation unit 38 calculates a luminance difference with respect to the bird's-eye view image data subjected to the viewpoint conversion by the viewpoint conversion unit 31 in order to detect the edge of the three-dimensional object included in the bird's-eye view image. For each of a plurality of positions along a vertical imaginary line extending in the vertical direction in the real space, the brightness difference calculating unit 38 calculates a brightness difference between two pixels in the vicinity of each position. The luminance difference calculation unit 38 can calculate the luminance difference by either a method of setting only one vertical virtual line extending in the vertical direction in the real space or a method of setting two vertical virtual lines.

  Here, a specific method for setting two vertical virtual lines will be described. The brightness difference calculation unit 38 applies a first vertical imaginary line corresponding to a line segment extending in the vertical direction in the real space and a vertical direction in the real space different from the first vertical imaginary line with respect to the bird's eye view image that has undergone viewpoint conversion. A second vertical imaginary line corresponding to the extending line segment is set. The luminance difference calculation unit 38 continuously calculates the luminance difference between the points on the first vertical imaginary line and the points on the second vertical imaginary line along the first vertical imaginary line and the second vertical imaginary line. Hereinafter, the operation of the luminance difference calculation unit 38 will be described in detail.

  As shown in FIG. 18A, the luminance difference calculation unit 38 corresponds to a line segment extending in the vertical direction in real space, and passes through the detection area A1 (hereinafter, attention line La). Set). In addition, unlike the attention line La, the luminance difference calculation unit 38 corresponds to a line segment extending in the vertical direction in the real space and also passes through the second vertical virtual line Lr (hereinafter referred to as a reference line Lr) passing through the detection area A1. Set. Here, the reference line Lr is set at a position separated from the attention line La by a predetermined distance in the real space. Note that the line corresponding to the line segment extending in the vertical direction in the real space is a line that spreads radially from the position Ps of the camera 10 in the bird's-eye view image. This radially extending line is a line along the direction in which the three-dimensional object falls when converted to bird's-eye view.

  The luminance difference calculation unit 38 sets a point of interest Pa (a point on the first vertical imaginary line) on the line of interest La. The luminance difference calculation unit 38 sets a reference point Pr (a point on the second vertical plate) on the reference line Lr. The attention line La, the attention point Pa, the reference line Lr, and the reference point Pr have the relationship shown in FIG. 18B in the real space. As apparent from FIG. 18B, the attention line La and the reference line Lr are lines extending in the vertical direction in the real space, and the attention point Pa and the reference point Pr are substantially the same height in the real space. This is the point that is set. Note that the attention point Pa and the reference point Pr do not necessarily have the same height, and an error that allows the attention point Pa and the reference point Pr to be regarded as the same height is allowed.

  The luminance difference calculation unit 38 calculates the luminance difference between the attention point Pa and the reference point Pr. If the luminance difference between the attention point Pa and the reference point Pr is large, it is considered that an edge exists between the attention point Pa and the reference point Pr. In particular, in the second embodiment, in order to detect a three-dimensional object existing in the detection areas A1 and A2, a vertical virtual line is set as a line segment extending in the vertical direction in the real space with respect to the bird's-eye view image, In the case where the luminance difference between the attention line La and the reference line Lr is high, there is a high possibility that there is an edge of the three-dimensional object at the set position of the attention line La. For this reason, the edge line detection unit 39 shown in FIG. 16 detects an edge line based on the luminance difference between the attention point Pa and the reference point Pr.

  This point will be described in more detail. FIG. 19 is a diagram illustrating a detailed operation of the luminance difference calculation unit 38, FIG. 19 (a) shows a bird's-eye view image in a bird's-eye view state, and FIG. 19 (b) shows FIG. 19 (a). It is the figure which expanded a part B1 of the bird's-eye view image. Although only the detection area A1 is illustrated and described in FIG. 19, the luminance difference is calculated in the same procedure for the detection area A2.

  When the adjacent vehicle V2 is reflected in the captured image captured by the camera 10, the adjacent vehicle V2 appears in the detection area A1 in the bird's-eye view image as shown in FIG. As shown in the enlarged view of the region B1 in FIG. 19A in FIG. 19B, it is assumed that the attention line La is set on the rubber part of the tire of the adjacent vehicle V2 on the bird's-eye view image. In this state, the luminance difference calculation unit 38 first sets a reference line Lr. The reference line Lr is set along the vertical direction at a position away from the attention line La by a predetermined distance in the real space. Specifically, in the three-dimensional object detection device 1a according to the present embodiment, the reference line Lr is set at a position separated from the attention line La by 10 cm in the real space. Thereby, the reference line Lr is set on the wheel of the tire of the adjacent vehicle V2, which is separated from the rubber of the tire of the adjacent vehicle V2, for example, by 10 cm, on the bird's eye view image.

  Next, the luminance difference calculation unit 38 sets a plurality of attention points Pa1 to PaN on the attention line La. In FIG. 19B, for the convenience of explanation, six attention points Pa1 to Pa6 (hereinafter simply referred to as attention points Pai when showing arbitrary points) are set. Note that the number of attention points Pa set on the attention line La may be arbitrary. In the following description, it is assumed that N attention points Pa are set on the attention line La.

  Next, the luminance difference calculation unit 38 sets the reference points Pr1 to PrN so as to be the same height as the attention points Pa1 to PaN in the real space. Then, the luminance difference calculation unit 38 calculates the luminance difference between the attention point Pa and the reference point Pr having the same height. Thereby, the brightness | luminance difference calculation part 38 calculates the brightness | luminance difference of two pixels for every some position (1-N) along the vertical virtual line extended in the perpendicular direction in real space. For example, the luminance difference calculation unit 38 calculates a luminance difference between the first point of interest Pa1 and the first reference point Pr1, and the luminance difference between the second point of interest Pa2 and the second reference point Pr2. Will be calculated. Thereby, the luminance difference calculation unit 38 continuously calculates the luminance difference along the attention line La and the reference line Lr. That is, the luminance difference calculation unit 38 sequentially obtains the luminance difference between the third to Nth attention points Pa3 to PaN and the third to Nth reference points Pr3 to PrN.

  The luminance difference calculation unit 38 repeatedly executes the processing such as setting the reference line Lr, setting the attention point Pa and the reference point Pr, and calculating the luminance difference while shifting the attention line La in the detection area A1. That is, the luminance difference calculation unit 38 repeatedly executes the above processing while changing the position of each of the attention line La and the reference line Lr by the same distance in the extending direction of the ground line L1 in the real space. For example, the luminance difference calculation unit 38 sets the reference line Lr as the reference line Lr in the previous process, sets the reference line Lr for the attention line La, and sequentially obtains the luminance difference. It will be.

  As described above, in the second embodiment, the edge extending in the vertical direction is obtained by calculating the luminance difference from the attention point Pa on the attention line La and the reference point Pr on the reference line Lr that are substantially the same height in the real space. It is possible to clearly detect a luminance difference in the case where there is. Also, in order to compare the brightness of vertical virtual lines extending in the vertical direction in real space, even if the three-dimensional object is stretched according to the height from the road surface by converting to a bird's-eye view image, The detection process is not affected, and the detection accuracy of the three-dimensional object can be improved.

  Returning to FIG. 16, the edge line detection unit 39 detects an edge line from the continuous luminance difference calculated by the luminance difference calculation unit 38. For example, in the case illustrated in FIG. 19B, the first attention point Pa <b> 1 and the first reference point Pr <b> 1 are located in the same tire portion, and thus the luminance difference is small. On the other hand, the second to sixth attention points Pa2 to Pa6 are located in the rubber part of the tire, and the second to sixth reference points Pr2 to Pr6 are located in the wheel part of the tire. Therefore, the luminance difference between the second to sixth attention points Pa2 to Pa6 and the second to sixth reference points Pr2 to Pr6 increases. For this reason, the edge line detection unit 39 can detect that an edge line exists between the second to sixth attention points Pa2 to Pa6 and the second to sixth reference points Pr2 to Pr6 having a large luminance difference. it can.

Specifically, when detecting the edge line, the edge line detection unit 39 firstly follows the following equation 1 to determine the i-th attention point Pai (coordinates (xi, yi)) and the i-th reference point Pri (coordinates (xi). ', Yi')), the i th attention point Pai is attributed.
[Formula 1]
When I (xi, yi)> I (xi ′, yi ′) + t s (xi, yi) = 1
When I (xi, yi) <I (xi ′, yi ′) − t s (xi, yi) = − 1
Otherwise s (xi, yi) = 0

  In Equation 1, t represents an edge threshold, I (xi, yi) represents the luminance value of the i-th attention point Pai, and I (xi ′, yi ′) represents the luminance value of the i-th reference point Pri. Show. According to the above equation 1, when the luminance value of the attention point Pai is higher than the luminance value obtained by adding the threshold value t to the reference point Pri, the attribute s (xi, yi) of the attention point Pai is “1”. Become. On the other hand, when the luminance value of the attention point Pai is lower than the luminance value obtained by subtracting the edge threshold t from the reference point Pri, the attribute s (xi, yi) of the attention point Pai is “−1”. When the luminance value of the attention point Pai and the luminance value of the reference point Pri are in other relationships, the attribute s (xi, yi) of the attention point Pai is “0”. In the present embodiment, the edge threshold value t may be changed by a threshold value changing unit 37a described later. When the edge threshold value t is changed by the threshold value changing unit 37a, the edge threshold value t is changed by the threshold value changing unit 37a. Using the edge threshold t, the attribute s (xi, yi) of the attention point Pai is detected.

Next, the edge line detection unit 39 determines whether the attention line La is an edge line from the continuity c (xi, yi) of the attribute s along the attention line La based on the following formula 2.
[Formula 2]
When s (xi, yi) = s (xi + 1, yi + 1) (and excluding 0 = 0),
c (xi, yi) = 1
Other than the above
c (xi, yi) = 0

  When the attribute s (xi, yi) of the attention point Pai and the attribute s (xi + 1, yi + 1) of the adjacent attention point Pai + 1 are the same, the continuity c (xi, yi) is ‘1’. When the attribute s (xi, yi) of the attention point Pai is not the same as the attribute s (xi + 1, yi + 1) of the adjacent attention point Pai + 1, the continuity c (xi, yi) is “0”.

  Next, the edge line detection unit 39 obtains the sum for the continuity c of all the attention points Pa on the attention line La. The edge line detection unit 39 normalizes the continuity c by dividing the obtained sum of the continuity c by the number N of points of interest Pa. Then, the edge line detection unit 39 determines that the attention line La is an edge line when the normalized value exceeds the threshold θ. The threshold value θ is a value set in advance through experiments or the like.

That is, the edge line detection unit 39 determines whether or not the attention line La is an edge line based on the following Equation 3. Then, the edge line detection unit 39 determines whether or not all the attention lines La drawn on the detection area A1 are edge lines.
[Formula 3]
Σc (xi, yi) / N> θ

  As described above, in the second embodiment, the attention point Pa is attributed based on the luminance difference between the attention point Pa on the attention line La and the reference point Pr on the reference line Lr, and the attribute along the attention line La is attributed. Since it is determined whether the attention line La is an edge line based on the continuity c of the image, the boundary between the high luminance area and the low luminance area is detected as an edge line, and an edge in line with a natural human sense Detection can be performed. This effect will be described in detail. FIG. 20 is a diagram illustrating an image example for explaining the processing of the edge line detection unit 39. In this image example, a first striped pattern 101 showing a striped pattern in which a high brightness area and a low brightness area are repeated, and a second striped pattern showing a striped pattern in which a low brightness area and a high brightness area are repeated. 102 is an adjacent image. Further, in this image example, a region where the brightness of the first striped pattern 101 is high and a region where the brightness of the second striped pattern 102 is low are adjacent to each other, and a region where the brightness of the first striped pattern 101 is low and the second striped pattern 102. Is adjacent to a region with high brightness. The portion 103 located at the boundary between the first striped pattern 101 and the second striped pattern 102 tends not to be perceived as an edge depending on human senses.

  On the other hand, since the low luminance region and the high luminance region are adjacent to each other, if the edge is detected only by the luminance difference, the part 103 is recognized as an edge. However, since the edge line detection unit 39 determines the part 103 as an edge line only when the attribute of the luminance difference has continuity in addition to the luminance difference in the part 103, the edge line detection unit 39 An erroneous determination of recognizing a part 103 that is not recognized as an edge line as a sensation as an edge line can be suppressed, and edge detection according to a human sense can be performed.

  Returning to FIG. 16, the three-dimensional object detection unit 33 a detects a three-dimensional object based on the amount of edge lines detected by the edge line detection unit 39. As described above, the three-dimensional object detection device 1a according to the present embodiment detects an edge line extending in the vertical direction in real space. The fact that many edge lines extending in the vertical direction are detected means that there is a high possibility that a three-dimensional object exists in the detection areas A1 and A2. For this reason, the three-dimensional object detection unit 33a detects a three-dimensional object based on the amount of edge lines detected by the edge line detection unit 39. Specifically, the three-dimensional object detection unit 33a determines whether or not the amount of edge lines detected by the edge line detection unit 39 is equal to or greater than a predetermined threshold value β, and the amount of edge lines is determined to be a predetermined threshold value β. In the case described above, the edge line detected by the edge line detection unit 39 is determined to be an edge line of a three-dimensional object.

  Further, prior to detecting the three-dimensional object, the three-dimensional object detection unit 33a determines whether or not the edge line detected by the edge line detection unit 39 is correct. The three-dimensional object detection unit 33a determines whether or not the luminance change along the edge line of the bird's-eye view image on the edge line is equal to or greater than a predetermined threshold value tb. When the brightness change of the bird's-eye view image on the edge line is equal to or greater than the threshold value tb, it is determined that the edge line has been detected by erroneous determination. On the other hand, when the luminance change of the bird's-eye view image on the edge line is less than the threshold value tb, it is determined that the edge line is correct. The threshold value tb is a value set in advance by experiments or the like.

  FIG. 21 is a diagram illustrating the luminance distribution of the edge line. FIG. 21A illustrates the edge line and the luminance distribution when the adjacent vehicle V2 as a three-dimensional object exists in the detection area A1, and FIG. Indicates an edge line and a luminance distribution when there is no solid object in the detection area A1.

  As shown in FIG. 21A, it is assumed that the attention line La set in the tire rubber portion of the adjacent vehicle V2 is determined to be an edge line in the bird's-eye view image. In this case, the luminance change of the bird's-eye view image on the attention line La is gentle. This is because the tire of the adjacent vehicle is extended in the bird's-eye view image by converting the image captured by the camera 10 into the bird's-eye view image. On the other hand, as shown in FIG. 21B, it is assumed that the attention line La set in the white character portion “50” drawn on the road surface in the bird's-eye view image is erroneously determined as an edge line. In this case, the brightness change of the bird's-eye view image on the attention line La has a large undulation. This is because a portion with high brightness in white characters and a portion with low brightness such as a road surface are mixed on the edge line.

  Based on the difference in luminance distribution on the attention line La as described above, the three-dimensional object detection unit 33a determines whether or not the edge line is detected by erroneous determination. For example, when a captured image acquired by the camera 10 is converted into a bird's-eye view image, the three-dimensional object included in the captured image tends to appear in the bird's-eye view image in a stretched state. As described above, when the tire of the adjacent vehicle V2 is stretched, one portion of the tire is stretched, so that the luminance change of the bird's eye view image in the stretched direction tends to be small. On the other hand, when a character or the like drawn on the road surface is erroneously determined as an edge line, the bird's-eye view image includes a high luminance region such as a character portion and a low luminance region such as a road surface portion. In this case, the brightness change in the stretched direction tends to increase in the bird's-eye view image. Therefore, when the luminance change along the edge line is greater than or equal to the predetermined threshold value tb, the three-dimensional object detection unit 33a detects the edge line by erroneous determination, and the edge line is detected by the three-dimensional object. Judge that it is not caused. Thereby, white characters such as “50” on the road surface, weeds on the road shoulder, and the like are determined as edge lines, and the detection accuracy of the three-dimensional object is prevented from being lowered. On the other hand, when the change in luminance along the edge line is less than the predetermined threshold value tb, the three-dimensional object detection unit 33a determines that the edge line is an edge line of the three-dimensional object, and the three-dimensional object exists. Judge.

Specifically, the three-dimensional object detection unit 33a calculates the luminance change of the edge line according to any of the following formulas 4 and 5. The luminance change of the edge line corresponds to the evaluation value in the vertical direction in the real space. Equation 4 below evaluates the luminance distribution by the sum of the squares of the differences between the i-th luminance value I (xi, yi) on the attention line La and the adjacent i + 1-th luminance value I (xi + 1, yi + 1). . Equation 5 below evaluates the luminance distribution by the sum of the absolute values of the differences between the i-th luminance value I (xi, yi) on the attention line La and the adjacent i + 1-th luminance value I (xi + 1, yi + 1). To do.
[Formula 4]
Evaluation value in the vertical equivalent direction = Σ [{I (xi, yi) −I (xi + 1, yi + 1)} ² ]
[Formula 5]
Evaluation value in the vertical equivalent direction = Σ | I (xi, yi) −I (xi + 1, yi + 1) |

Not only the above formula 5, but also the attribute b of the adjacent luminance value is binarized using the threshold value t2 as in the following formula 6, and the binarized attribute b is summed for all the attention points Pa. May be.
[Formula 6]
Evaluation value in the vertical equivalent direction = Σb (xi, yi)
However, when | I (xi, yi) −I (xi + 1, yi + 1) |> t2,
b (xi, yi) = 1
Other than the above
b (xi, yi) = 0

  When the absolute value of the luminance difference between the luminance value of the attention point Pai and the luminance value of the reference point Pri is larger than the threshold value t2, the attribute b (xi, yi) of the attention point Pa (xi, yi) is “1”. Become. If the relationship is other than that, the attribute b (xi, yi) of the attention point Pai is '0'. This threshold value t2 is set in advance by an experiment or the like in order to determine that the attention line La is not on the same three-dimensional object. Then, the three-dimensional object detection unit 33a sums the attributes b for all the attention points Pa on the attention line La and obtains an evaluation value in the vertical equivalent direction, whereby the edge line is caused by the three-dimensional object. It is determined whether or not a three-dimensional object exists.

  Returning to FIG. 16, the threshold value changing unit 37 a performs imaging based on the white turbidity of the lens calculated by the white turbidity calculating unit 34 and the reliability of the headlight of the adjacent vehicle V <b> 2 calculated by the certainty calculation unit 36. The edge threshold t for detecting the edge of the three-dimensional object from the image is changed. Specifically, as in the first embodiment, the threshold value changing unit 37a increases the certainty that the light source detected by the headlight detection unit 35 is the headlight of the adjacent vehicle V2, so It is determined that the adjacent vehicle V2 exists, and the edge threshold value t is changed to a low value so that a three-dimensional object (adjacent vehicle V2) existing in the adjacent lane is easily detected. Further, the threshold value changing unit 37a increases the difference threshold th and the headlight certainty of the adjacent vehicle V2 such that the higher the lens turbidity, the lower the difference threshold th relative to the headlight certainty of the adjacent vehicle V2. Change the map to show the relationship. Accordingly, even when the lens is clouded and the certainty that the detected light source is the headlight of the adjacent vehicle V2 is low, the light source detected by the headlight detection unit 35 is the adjacent vehicle V2. It is possible to appropriately determine whether or not the headlight is, and as a result, it is possible to appropriately detect the adjacent vehicle V2.

  Next, an adjacent vehicle detection method according to the second embodiment will be described. FIG. 22 is a flowchart showing details of the adjacent vehicle detection method according to the second embodiment. In FIG. 22, for the sake of convenience, the process for the detection area A1 will be described, but the same process is executed for the detection area A2.

  In step S301, the camera 10 captures a predetermined area specified by the angle of view a and the attachment position, and the computer 30a acquires image data of the captured image P captured by the camera 10. Next, in step S302, the viewpoint conversion unit 31 performs viewpoint conversion on the acquired image data to generate bird's-eye view image data.

  Next, in step S303, the luminance difference calculation unit 38 sets the attention line La on the detection area A1. At this time, the luminance difference calculation unit 38 sets a line corresponding to a line extending in the vertical direction in the real space as the attention line La. Next, the luminance difference calculation unit 38 sets a reference line Lr on the detection area A1 in step S304. At this time, the luminance difference calculation unit 38 sets a reference line Lr that corresponds to a line segment extending in the vertical direction in the real space and is separated from the attention line La by a predetermined distance in the real space.

  Next, in step S305, the luminance difference calculation unit 38 sets a plurality of attention points Pa on the attention line La. At this time, the luminance difference calculation unit 38 sets the attention points Pa as many as not causing a problem when the edge detection unit 39 detects the edge. In step S306, the luminance difference calculation unit 38 sets the reference point Pr so that the attention point Pa and the reference point Pr are substantially the same height in the real space. Thereby, the attention point Pa and the reference point Pr are arranged in a substantially horizontal direction, and it becomes easy to detect an edge line extending in the vertical direction in the real space.

  Next, in step S307, the luminance difference calculation unit 38 calculates the luminance difference between the attention point Pa and the reference point Pr, which have the same height in the real space. Then, the edge line detection unit 39 calculates the attribute s of each attention point Pa based on the luminance difference calculated by the luminance difference calculation unit 38 according to the above equation 1. In the present embodiment, the attribute s of each attention point Pa is calculated using the edge threshold value t for detecting the edge of the adjacent vehicle V2. The edge threshold value t may be changed in a threshold value changing process described later. When the edge threshold value t is changed, the changed edge threshold value t is used in this step S307.

  Next, in step S308, the edge line detection unit 39 calculates the continuity c of the attribute s of each attention point Pa according to the above equation 2. In step S309, the edge line detection unit 39 determines whether the value obtained by normalizing the total sum of continuity c is greater than the threshold value θ according to the above equation 3. If it is determined that the normalized value is larger than the threshold θ (step S309 = Yes), the edge line detection unit 39 detects the attention line La as an edge line in step S310. Then, the process proceeds to step S311. When it is determined that the normalized value is not larger than the threshold θ (step S309 = No), the edge line detection unit 39 does not detect the attention line La as an edge line, and the process proceeds to step S311.

  In step S311, the computer 30a determines whether or not the processing in steps S303 to S310 has been executed for all the attention lines La that can be set on the detection area A1. When it is determined that the above processing has not been performed for all the attention lines La (step S311 = No), the processing returns to step S303, a new attention line La is set, and the processing up to step S311 is repeated. On the other hand, if it is determined that the above process has been performed for all the attention lines La (step S311 = Yes), the process proceeds to step S312.

  In step S312, the three-dimensional object detection unit 33a calculates a luminance change along the edge line for each edge line detected in step S310. The three-dimensional object detection unit 33a calculates the luminance change of the edge line according to any one of the above formulas 4, 5, and 6. Next, in step S313, the three-dimensional object detection unit 33a excludes edge lines whose luminance change is equal to or greater than a predetermined threshold value tb from among the edge lines. That is, it is determined that an edge line having a large luminance change is not a correct edge line, and the edge line is not used for detecting a three-dimensional object. As described above, this is to prevent characters on the road surface, roadside weeds, and the like included in the detection area A1 from being detected as edge lines. Therefore, the predetermined threshold value tb is a value set based on a luminance change generated by characters on the road surface, weeds on the road shoulder, or the like, which is obtained in advance through experiments or the like. On the other hand, the three-dimensional object detection unit 33a determines an edge line whose luminance change is less than the predetermined threshold value tb among the edge lines as an edge line of the three-dimensional object, and thereby detects a three-dimensional object existing in the adjacent vehicle. .

  Next, in step S314, the three-dimensional object detection unit 33a determines whether or not the amount of the edge line is equal to or greater than a predetermined threshold value β. Here, the threshold value β is a value set in advance by experiments or the like. For example, when a four-wheeled vehicle is set as a three-dimensional object to be detected, the threshold value β is determined in advance by an experiment or the like. It is set based on the number of edge lines of the four-wheeled vehicle that appeared in A1. When it is determined that the amount of the edge line is equal to or larger than the threshold value β (step S314 = Yes), the three-dimensional object detection unit 33a determines that a three-dimensional object exists in the detection area A1, and proceeds to step S315. It is determined that a vehicle exists. On the other hand, when it is determined that the amount of the edge line is not equal to or larger than the threshold β (step S314 = No), the three-dimensional object detection unit 33a determines that there is no three-dimensional object in the detection area A1, and proceeds to step S316. It is determined that there is no adjacent vehicle in the detection area A1.

  Subsequently, the threshold value changing process according to the second embodiment will be described with reference to FIG. Note that the threshold value changing process according to the second embodiment is also performed in parallel with the adjacent vehicle detection process shown in FIG. 22 as in the first embodiment. In addition, the threshold value changing process according to the second embodiment is a three-dimensional object so that the adjacent vehicle V2 can be appropriately detected even when foreign matters such as scale adhere to the lens and the lens is clouded. The edge threshold t for detecting is changed. Therefore, the edge threshold value t changed in this threshold value changing process is used when detecting the edge of the three-dimensional object (adjacent vehicle V2) existing in the adjacent lane in step S307 of the adjacent vehicle detection process shown in FIG. It becomes. FIG. 23 is a diagram for explaining the threshold value changing process according to the second embodiment.

  In steps S401 to S404 of the threshold value changing process according to the second embodiment, light sources that are candidates for headlights of the adjacent vehicle V2 are detected (step S401), as in steps S201 to S204 of the first embodiment. When a light source that is a candidate for a headlight of the adjacent vehicle V2 is detected (step S402 = Yes), a certainty factor is calculated that the detected light source is a headlight of the adjacent vehicle V2 (step S403). Subsequently, the lens turbidity is calculated (step S404). On the other hand, when the headlight of the adjacent vehicle V2 is not detected (step S402 = No), the detection of the headlight of the adjacent vehicle V2 is repeatedly performed.

  In step S405, the threshold changing unit 37a selects a map indicating the relationship between the headlight certainty of the adjacent vehicle V2 and the edge threshold t based on the turbidity of the lens calculated in step S404. Is called. Specifically, as in the first embodiment, the threshold change unit 37a sets the edge threshold th to be lower as the turbidity of the lens is higher, even when the certainty of the headlight of the adjacent vehicle V2 is low. The right map.

  In step S406, the threshold value changing unit 37a changes the edge threshold value t based on the certainty of the headlight of the adjacent vehicle V2 calculated in step S403 and the map selected in step S405. Specifically, the threshold value changing unit 37a changes the edge threshold value th to a lower value as the degree of certainty that the detected light source is the headlight of the adjacent vehicle V2 is higher according to the map selected in step S405. . Thereby, also in the second embodiment, even when the lens is clouded, if the certainty that the light source is the headlight of the adjacent vehicle V2 is high and it can be determined that the adjacent vehicle V2 exists, the adjacent lane The edge threshold t for detecting the three-dimensional object (adjacent vehicle V2) existing in the vehicle is changed to a low value, and as a result, the adjacent vehicle V2 existing in the adjacent lane can be appropriately detected. Then, after the edge threshold value t is changed in step S406, the process returns to step S401, and the threshold value changing process described above is repeatedly performed.

  As described above, in the second embodiment, when the adjacent vehicle V2 is detected based on the edge of the three-dimensional object existing in the adjacent lane, the certainty that the detected light source is the headlight of the adjacent vehicle V2. The higher the value is, the lower the edge threshold value t is. Thereby, when the certainty factor that the detected light source is the headlight of the adjacent vehicle V2 is high and it can be determined that the adjacent vehicle V2 exists, the three-dimensional object (adjacent vehicle V2) existing in the adjacent lane is It becomes easy to detect, and it is possible to appropriately detect the adjacent vehicle V2 even when foreign matters such as water scale adhere to the lens and the lens is clouded. In the second embodiment, the degree of white turbidity of the lens is calculated as white turbidity. The higher the white turbidity of the lens, the higher the certainty that the detected light source is the headlight of the adjacent vehicle V2. As described above, by changing the map of the reliability of the headlight of the adjacent vehicle V2 and the edge threshold value t, the lens is clouded, and the reliability of the detected light source is the headlight of the adjacent vehicle V2. Even when it becomes low, the headlight of the adjacent vehicle V2 can be appropriately detected, and as a result, the adjacent vehicle V2 can be appropriately detected.

  The embodiment described above is described for facilitating the understanding of the present invention, and is not described for limiting the present invention. Therefore, each element disclosed in the above embodiment is intended to include all design changes and equivalents belonging to the technical scope of the present invention.

  For example, in the above-described embodiment, the configuration in which the difference threshold th or the edge threshold t is changed to a lower value as the headlight reliability of the adjacent vehicle V2 is higher is exemplified. The higher the certainty of the headlight of the adjacent vehicle V2, the lower the threshold value α and the threshold value t2 for detecting an edge line. It is good also as a structure changed into a value. As a result, the higher the certainty of the headlight of the adjacent vehicle V2 is, the easier it is to detect the adjacent vehicle V2. Therefore, even when the lens is clouded, it is possible to appropriately detect the adjacent vehicle V2 existing in the adjacent lane. It becomes. Moreover, it is good also as a structure which makes the pixel value (or luminance value) output from the camera 10 high based on the certainty degree of the headlight of the adjacent vehicle V2. In this case, even if the difference threshold th and the edge threshold t are not changed, the difference pixel DP and the edge are easily detected, and the three-dimensional object (adjacent vehicle V2) is easily detected. It is possible to appropriately detect the adjacent vehicle V2 existing in the lane.

  Further, in the above-described embodiment, the configuration in which the three-dimensional object is detected as the adjacent vehicle V2 when the moving speed of the three-dimensional object satisfies the predetermined condition is illustrated, but the configuration is not limited to this configuration, and for example, the adjacent vehicle V2 It is good also as a structure which accelerates | stimulates the detection of the adjacent vehicle V2 by changing said conditions based on the certainty degree of the headlight. For example, in the above-described embodiment, when the absolute moving speed of the three-dimensional object is 10 km / h or more and the relative moving speed of the three-dimensional object with respect to the host vehicle V1 is +60 km / h or less, the three-dimensional object is determined as the adjacent vehicle V2. However, when the certainty of the headlight of the adjacent vehicle V2 is high, for example, the absolute moving speed of the three-dimensional object is 5 km / h or more, and the relative moving speed of the three-dimensional object with respect to the host vehicle V1 is +70 km / h. In the following cases, it can be determined that the three-dimensional object is the adjacent vehicle V2.

  In the above-described embodiment, the map indicating the relationship between the difference threshold th or the edge threshold t and the certainty of the headlights of the adjacent vehicle V2 according to the white turbidity of the lens calculated by the white turbidity calculation unit 34. Although the configuration to be changed is exemplified, the configuration is not limited to this configuration. For example, the higher the turbidity of the lens, the higher the reliability of the headlight calculated by the reliability calculation unit 36 may be used. In particular, the higher the white turbidity of the lens, the higher the adjacent vehicle so that the detected light source can be determined as the headlight of the adjacent vehicle V2 even when the size of the light source detected by the headlight detection unit 35 is small. It can be set as the structure which calculates highly the reliability of the headlight of V2.

  Furthermore, in the above-described embodiment, the higher the certainty of the headlight of the adjacent vehicle V2 and the higher the white turbidity of the lens, the easier it is to detect a three-dimensional object (adjacent vehicle V2) present in the adjacent lane. However, the present invention is not limited to this configuration. For example, the turbidity of the lens is not less than a predetermined value, and the certainty of the headlight of the adjacent vehicle V2 is predetermined. When the value is equal to or greater than the value, the detected light source may be a headlight of the adjacent vehicle V2, and it may be determined that the adjacent vehicle V2 exists in the adjacent lane. That is, unlike the above-described embodiment, when the headlight of the adjacent vehicle V2 is detected without considering the differential waveform or the like, it is determined that the adjacent vehicle V2 exists and the adjacent vehicle V2 is detected. It can be.

  In the above-described embodiment, the configuration for calculating the turbidity of the lens by calculating the sharpness of the image is exemplified. However, the configuration is not limited to this configuration. For example, the turbidity of the lens is set as follows. It is good also as a structure to calculate. That is, when a light source that is a candidate for a headlight of the adjacent vehicle V2 is detected, an edge line of a three-dimensional object existing in the adjacent lane is detected from the captured image, and based on the number of edge lines detected within a predetermined time. Thus, the configuration may be such that the turbidity of the lens is calculated. For example, when the headlight of the adjacent vehicle V2 is detected when the lens is not clouded, an edge line corresponding to the adjacent vehicle V2 can be detected. However, when the cloudiness of the lens is high In some cases, the image of the adjacent vehicle V2 is blurred and the edge line corresponding to the adjacent vehicle V2 cannot be detected. Therefore, if the number of edge lines detected within a predetermined time after detecting the headlight of the adjacent vehicle V2 is small, it can be determined that the lens is clouded.

Further, in addition to the above-described embodiment, the adjacent vehicle detection process based on the differential waveform DW _t shown in FIG. 14 and the adjacent vehicle detection process based on the edge shown in FIG. 22 may be performed in parallel. In this case, when the difference threshold th and the edge threshold t are changed based on the white turbidity of the lens detected by the white turbidity calculator 34 and the certainty of the headlight of the adjacent vehicle V2, first, the edge threshold t After changing to a certain value, the difference threshold th can be changed. When foreign matter such as scale adheres to the lens and the lens is clouded, the possibility of erroneously detecting the adjacent vehicle as the adjacent vehicle V2 is low even if the edge threshold value t is lowered. Therefore, first, the edge threshold value t is changed to a certain value, and then the difference threshold value th is changed, so that it is possible to effectively prevent the adjacent adjacent vehicle from being erroneously detected as the adjacent vehicle V2.

  Note that the camera 10 of the above-described embodiment corresponds to the imaging unit of the present invention, the viewpoint conversion unit 31 corresponds to the image conversion unit of the present invention, the alignment unit 32, the three-dimensional object detection units 33 and 33a, and the luminance difference calculation. The unit 38 and the edge line detection unit 39 correspond to the three-dimensional object detection unit of the present invention, the white turbidity calculation unit 34 corresponds to the white turbidity calculation unit of the present invention, and the headlight detection unit 35 corresponds to the light source detection unit of the present invention. The certainty factor calculation unit 36 corresponds to the certainty factor calculation unit of the present invention, and the threshold value changing units 37 and 37a correspond to the control unit of the present invention.

DESCRIPTION OF SYMBOLS 1, 1a ... Three-dimensional object detection apparatus 10 ... Camera 20 ... Vehicle speed sensor 30, 30a ... Computer 31 ... Viewpoint conversion part 32 ... Position alignment part 33 ... Three-dimensional object detection part 34 ... White turbidity calculation part 35 ... Headlight detection part 36 ... Certainty factor calculation units 37, 37a ... threshold change unit 38 ... luminance difference calculation unit 39 ... edge line detection unit a ... view angle A1, A2 ... detection region CP ... intersection DP ... difference pixel DW _t , DW _t '... difference waveform DW _{t1 ~DW m, DW m + k} ~DW tn ... small regions L1, L2 ... ground line La, Lb ... line P ... captured image on the direction in which the three-dimensional object collapses PB t _... bird's-eye view image PD t _... difference image V1 ... vehicle V2 ... Adjacent vehicle

Claims (17)

An imaging means comprising a lens for forming an image of the rear of the vehicle;
A three-dimensional object detecting means for detecting a three-dimensional object including an adjacent vehicle based on a captured image obtained by the imaging means;
Based on the captured image, a turbidity calculating means for calculating the degree of turbidity of the lens as the turbidity;
Light source detection means for detecting a light source present behind the host vehicle based on the captured image;
A certainty factor calculating unit that calculates a certainty factor that the light source is a headlight of the adjacent vehicle, based on an aspect of the light source detected by the light source detecting unit;
Control means for facilitating detection of the three-dimensional object by the three-dimensional object detection means, as the certainty level of the headlight of the adjacent vehicle increases.
The three-dimensional object detection device, wherein the control unit causes the certainty factor calculating unit to calculate the certainty factor of the headlight of the adjacent vehicle as the white turbidity of the lens is high. The three-dimensional object detection device according to claim 1,
The three-dimensional object detection means includes:
Image conversion means for converting the captured image obtained by the imaging means into a bird's-eye view image;
The position of the bird's-eye view image at different times obtained by the image conversion means is aligned on the bird's-eye view, and the number of pixels indicating a predetermined difference is counted on the difference image of the aligned bird's-eye view image. A three-dimensional object detection device, wherein differential waveform information is generated by distribution, and the three-dimensional object is detected based on the difference waveform information. The three-dimensional object detection device according to claim 2,
The three-dimensional object detection means generates the difference waveform information by counting the number of pixels showing a difference equal to or greater than a predetermined first threshold on the difference image, and generating the difference waveform information, and the peak value of the difference waveform information Is detected based on the differential waveform information when the value is equal to or greater than a predetermined second threshold,
The control means detects the three-dimensional object based on the difference waveform information by changing the first threshold value or the second threshold value to a lower value as the certainty of the headlight of the adjacent vehicle is higher. 3D object detection device characterized by promoting The three-dimensional object detection device according to claim 2 or 3,
When the three-dimensional object detection unit generates the difference waveform information, the control unit causes the three-dimensional object detection unit to give a predetermined difference on the difference image as the certainty of the headlight of the adjacent vehicle increases. A three-dimensional object detection device that promotes detection of the three-dimensional object based on the differential waveform information by counting the number of pixels to be displayed and outputting a high frequency distribution value. The three-dimensional object detection device according to claim 1,
The three-dimensional object detection means includes:
Image conversion means for converting the captured image obtained by the imaging means into a bird's-eye view image;
A three-dimensional object detection apparatus, wherein edge information is detected based on a bird's-eye view image obtained by the image conversion means, and the three-dimensional object is detected based on the edge information. The three-dimensional object detection device according to claim 5,
The three-dimensional object detection means detects an edge component whose luminance difference between adjacent pixel areas is not less than a predetermined first threshold value from the bird's eye view image, and the amount of the edge information based on the edge component is a predetermined second value. When the threshold is equal to or greater than the threshold, the solid object is detected based on the edge information,
The control means detects the three-dimensional object based on the edge information by changing the first threshold value or the second threshold value to a lower value as the certainty of the headlight of the adjacent vehicle is higher. A three-dimensional object detection device characterized by promoting. The three-dimensional object detection device according to claim 5 or 6,
When the solid object detection unit detects the edge information, the control unit causes the solid object detection unit to output the edge information higher as the certainty of the headlight of the adjacent vehicle increases. A three-dimensional object detection device that facilitates detecting the three-dimensional object based on the edge information. The three-dimensional object detection device according to any one of claims 1 to 7,
The certainty factor calculating means specifies the size of the light source detected by the light source detecting means as an aspect of the light source, and the reliability of the headlight of the adjacent vehicle increases as the specified light source size increases. A three-dimensional object detection device characterized by high calculation. The three-dimensional object detection device according to any one of claims 1 to 8,
The certainty factor calculating means identifies a time change of the position of the light source detected by the light source detection means as an aspect of the light source, and determines a moving speed of the light source based on the time change of the specified position of the light source. The three-dimensional object detection apparatus is characterized in that the higher the moving speed of the light source is, the higher the certainty of the headlight of the adjacent vehicle is calculated. The three-dimensional object detection device according to claim 1,
The certainty factor calculation means specifies the position of the light source detected by the light source detection means as an aspect of the light source, and the shorter the distance in the vehicle width direction from the specified light source to the host vehicle, A three-dimensional object detection device characterized by calculating a high degree of certainty of a headlight. The three-dimensional object detection device according to any one of claims 1 to 10,
The three-dimensional object detection device, wherein the white turbidity calculating means calculates the white turbidity of the lens based on a luminance gradient of a light source existing behind the host vehicle. The three-dimensional object detection device according to claim 1,
The three-dimensional object detection device, wherein the white turbidity calculating means calculates the white turbidity of the lens based on a specific frequency component obtained from the captured image. A three-dimensional object detection device according to any one of claims 1 to 12,
The three-dimensional object detection device, wherein the white turbidity calculating means calculates the white turbidity of the lens based on the intensity of edge information detected from the captured image. A captured image obtained by imaging the rear of the host vehicle is converted into a bird's-eye view image, difference waveform information is generated from a difference image of the bird's-eye view image at different times, a three-dimensional object is detected based on the difference waveform information, and the three-dimensional object A solid object detection method for determining whether an object is another vehicle,
A light source existing behind the host vehicle is detected based on the captured image, and a certainty factor that the light source is a headlight of the other vehicle is calculated based on the detected aspect of the light source. Based on this, the degree of cloudiness of the lens is calculated as white turbidity, and when calculating the certainty factor, the higher the white turbidity of the lens, the higher the certainty factor of the headlight of the other vehicle, The three-dimensional object is characterized in that the higher the certainty of the headlight of the vehicle, the more the detection of the three-dimensional object based on the difference waveform information or the determination of the three-dimensional object as the other vehicle is promoted. Object detection method. A captured image obtained by imaging the rear of the host vehicle is converted into a bird's-eye view image, edge information is detected based on the bird's-eye view image, a three-dimensional object is detected based on the edge information, and the three-dimensional object is another vehicle. A solid object detection method for determining whether or not
A light source existing behind the host vehicle is detected based on the captured image, and a certainty factor that the light source is a headlight of the other vehicle is calculated based on the detected aspect of the light source. Based on this, the degree of cloudiness of the lens is calculated as white turbidity, and when calculating the certainty factor, the higher the white turbidity of the lens, the higher the certainty factor of the headlight of the other vehicle, A three-dimensional object that promotes the detection of the three-dimensional object based on the edge information or the determination that the three-dimensional object is the other vehicle as the certainty of the headlight of the vehicle increases. Detection method. An imaging means comprising a lens for forming an image of the rear of the vehicle;
A three-dimensional object detecting means for detecting a three-dimensional object including an adjacent vehicle based on a captured image obtained by the imaging means;
Cloudiness determination means for determining whether the lens is clouded based on the captured image;
Light source detection means for detecting a light source present behind the host vehicle based on the captured image;
Headlight determination means for determining whether the light source is a headlight of the adjacent vehicle based on the aspect of the light source detected by the light source detection means;
Control means for facilitating detection of the three-dimensional object by the three-dimensional object detection means when it is determined that the light source is a headlight of the adjacent vehicle;
The control means makes it easier for the headlight determination means to determine that the light source is a headlight of the adjacent vehicle when it is determined that the lens is clouded. Detection device. A captured image obtained by imaging the rear of the host vehicle is converted into a bird's-eye view image, difference waveform information is generated from a difference image of the bird's-eye view image at different times, a three-dimensional object is detected based on the difference waveform information, and the three-dimensional object A solid object detection method for determining whether an object is another vehicle,
A light source existing behind the host vehicle is detected based on the captured image, and it is determined whether the light source is a headlight of the other vehicle based on the detected aspect of the light source. To determine whether the lens is cloudy,
In determining whether the light source is a headlight of the other vehicle, if it is determined that the lens is clouded, it is determined that the light source is a headlight of the other vehicle. When it is determined that the light source is a headlight of the other vehicle, the solid object is detected based on the difference waveform information, or the solid object is determined to be the other vehicle. A three-dimensional object detection method characterized by promoting the operation.

Images