JP2016081108A - Object detection device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an object detection device capable of reducing a load of image processing while suppressing a deterioration in object detection accuracy in object detection using an image.SOLUTION: An object detection device 1 for using an image picked up by a camera 2 mounted on a vehicle to detect an object around the vehicle acquires depth distance information in each pixel region constituting an image (S12), groups a plurality of pixel regions by associating the plurality of pixel regions on the basis of the depth distance information (S14), acquires optical flow 85 of an object to be picked up on the image in each pixel region to the image (S18), associates the optical flow 85 (S20) while the nearer the depth distance of a group of pixel regions is to the grouped image regions, the less the number of associations of the optical flow 85 is reduced, and detects a moving state of an object around the vehicle by using the associated optical flow 85 (S22).SELECTED DRAWING: Figure 7

Description

  The present invention relates to an object detection device.

  Conventionally, with respect to object detection, as described in US Pat. No. 8,548,229, a technique for generating a parallax image by performing stereo image processing on a captured image and generating a depth map based on the parallax image is known. It has been. This object detection technology groups pixels close to each other in a depth map into multiple segments, acquires movement information for each segment by optical flow, etc., and detects objects around the vehicle based on such segment information. The time until the object collides is to be estimated.

U.S. Pat. No. 8,548,229

  In the object detection technique described above, a detailed method for associating movement information with segments grouped based on distance information is not described. However, as this association process, each pixel constituting the segment is moved. It is possible to associate information.

  When performing the association process in this way, if the movement information is associated with each pixel, the load of the association process increases. On the other hand, it is conceivable that one piece of movement information is thinned out and associated with a plurality of pixels, but there is a concern that the object detection accuracy is lowered because movement information for the entire image is reduced.

  Therefore, in this technical field, it is desired to develop an object detection apparatus that can reduce the load of object detection processing while suppressing a decrease in object detection accuracy.

  That is, the object detection device according to one aspect of the present invention is an object detection device that detects an object around the vehicle using an image captured by a camera mounted on the vehicle, and for each pixel region constituting the image. A distance information acquisition unit that acquires depth distance information including a depth distance in the image, a grouping processing unit that groups the pixel regions based on the depth distance information, and for each pixel region with respect to the image In addition, a movement information acquisition unit for acquiring an optical flow including information on the movement direction and movement amount of the imaging target on the image, and the movement information acquisition for the pixel areas grouped by the grouping processing unit An association unit that associates the optical flow acquired by the association unit, and the optical flow associated by the association unit An object detecting unit that detects a moving state of an object around the vehicle using the optical flow, and the associating unit includes the optical flow per unit area of the image as the group of the pixel regions having a shorter depth distance. The number of associations is reduced, and the optical flow is associated.

  According to such an apparatus, optical flow association is performed by reducing the number of optical flow associations per unit area of an image for a group of pixel regions having a shorter depth distance. Thereby, the load of image processing is reduced as the group of pixel regions has a shorter depth distance. At this time, since the group of pixel areas with a close depth distance is displayed with a large object on the image, the detection accuracy of the object is not greatly reduced even if the number of optical flow associations is reduced. Accordingly, it is possible to reduce the load of the object detection process while suppressing a decrease in the object detection accuracy in the group of pixel regions having a short depth distance.

  In this object detection device, the associating unit may associate the optical flow with the pixel region so that the optical flows are associated with each other at a predetermined interval in the real space. In this case, by associating the optical flow on the image so that the optical flows are associated with each other at a preset interval in the real space, the optical flow corresponding to the object becomes closer as the depth distance of the imaged object becomes closer. The number of attachments is reduced. As a result, it is possible to reduce the load of the object detection process while suppressing a decrease in detection accuracy of an object with a short depth distance.

  ADVANTAGE OF THE INVENTION According to this invention, in the object detection using an image, the load of an object detection process can be reduced, suppressing the fall of object detection accuracy.

It is a block diagram which shows the structure outline | summary of the object detection apparatus which concerns on one Embodiment of this invention. It is explanatory drawing of the image used for the object detection of the object detection apparatus of FIG. It is explanatory drawing about the grouping process in the object detection apparatus of FIG. It is explanatory drawing about acquisition of the optical flow in the object detection apparatus of FIG. It is explanatory drawing regarding the matching of the optical flow in the object detection apparatus of FIG. It is explanatory drawing regarding the matching of the optical flow in the object detection apparatus of FIG. It is a flowchart of the object detection process in the object detection apparatus of FIG.

  Embodiments of the present invention will be described below with reference to the drawings. In the following description, the same elements are denoted by the same reference numerals, and redundant description is omitted.

  FIG. 1 is a schematic configuration diagram of an object detection apparatus 1 according to an embodiment of the present invention. FIG. 2 is an explanatory diagram of an image used for object detection of the object detection apparatus 1.

  As shown in FIG. 1, the object detection device 1 is a device that detects an object around the vehicle using an image captured by a camera 2 mounted on the vehicle. The object detection device 1 is mounted on a vehicle, for example, and includes a camera 2 and an ECU (Electronic Control Unit) 3.

  The camera 2 functions as an imaging unit that images the surroundings of the vehicle, and is attached, for example, so that the surroundings in the traveling direction of the vehicle can be captured. As the camera 2, a camera that can acquire luminance information and depth distance information of an image is used. That is, as the camera 2, a camera that can acquire luminance information for each pixel area constituting the image and depth distance information for each pixel area is used. For example, a stereo camera is used. In this case, the camera 2 having a plurality of imaging units arranged in a direction intersecting the imaging direction is used. Details of the pixel region will be described later.

  The image captured by the camera 2 may be color or monochrome. The imaging wavelength band may be a visible wavelength or a near-infrared wavelength, and may be any wavelength band as long as image data capable of recognizing an object around the vehicle is obtained. The camera 2 may be a sensor other than a stereo camera as long as it can acquire image luminance information and depth distance information. For example, a TOF (Time of Flight) camera may be used. A combination of a camera that captures an image of information and a sensor that acquires depth distance information, such as a laser radar sensor, may be used.

  The ECU 3 is an electronic control unit that controls the entire object detection apparatus 1 and is configured mainly by a computer including a CPU, a ROM, and a RAM, for example. The ECU 3 is electrically connected to the camera 2 and receives an imaging signal of the camera 2. The ECU 3 includes a distance information acquisition unit 30, a grouping processing unit 31, a movement information acquisition unit 32, an association unit 33, and an object detection unit 34.

  The distance information acquisition unit 30 acquires depth distance information in an image captured by the camera 2. For example, the distance information acquisition unit 30 acquires image data captured by the camera 2, and acquires depth distance information for each pixel region constituting the image using the image data. The image data is data that can acquire luminance information and depth distance information for each pixel region of the image, is repeatedly acquired at a predetermined cycle, and is recorded for each acquisition.

  The pixel area constituting the image may be an area composed of one pixel or an area composed of a plurality of pixels. When the pixel region is set as a region composed of a plurality of pixels, for example, a region composed of 4 × 2 × 2 pixels and a region composed of a larger number of pixels are used. The luminance information is, for example, the luminance value of the pixel area. When the pixel area is composed of one pixel, the luminance value of the pixel becomes the luminance information of the pixel area. When the pixel area is composed of a plurality of pixels, the average value, maximum value, minimum value, or predetermined representative value of the luminance values in the plurality of pixels is used as the luminance information of the pixel area. The depth distance information is information related to the depth distance of the imaging target imaged in the pixel area, and is information including the depth distance. For example, the depth distance is a distance to the imaging target in the imaging direction of the camera 2. When a stereo camera is used as the camera 2, the depth distance is a distance from a straight line connecting the installation positions of a plurality of cameras to the object to be imaged. The imaging target is captured in an image, and includes a background such as a road surface and sky in addition to an object that is a three-dimensional object. When the pixel area is composed of one pixel, the depth distance of the pixel is the depth distance information of the pixel area. When the pixel region is composed of a plurality of pixels, the average value, maximum value, minimum value, or predetermined representative value of the depth distance in the plurality of pixels may be used as the depth distance information of the pixel region.

  The depth distance information may use the value of the depth distance as it is, or may be a value corresponding to the depth distance. For example, a parallax value corresponding to the depth distance may be used as the depth distance information. The parallax value is the parallax value of the imaging object or object in the two images picked up by the two cameras, and becomes larger as the depth distance is shorter.

  When a stereo camera is used as the camera 2, the distance information acquisition unit 30 generates parallax images for the two left and right images. And the distance information acquisition part 30 uses the value of the parallax of the pixel area | region in a parallax image as depth distance information. Further, the depth distance for each pixel area may be calculated based on the parallax for each pixel area, and the value of the depth distance may be used as the depth distance information. As a method for calculating the depth distance information, for example, a dense stereo technique is used, and more specifically, an SGM method, an ELAS method, or the like is used.

  The grouping processing unit 31 groups a plurality of pixel regions based on the depth distance information for the image for which the depth distance information has been acquired. The grouping processing unit 31 sets adjacent pixel regions in the three-dimensional space as the same group and groups them. For example, the grouping processing unit 31 performs grouping as the same group when pixel regions within the same depth distance range are close to each other in the vertical direction of the image. The same depth distance range is a range including substantially the same distance within a predetermined depth distance range with respect to the same depth distance and the same depth distance. Further, the grouping processing unit 31 may perform grouping as the same group when pixel regions having the same depth distance range are close to each other in the vertical direction and the horizontal direction of the image.

  The grouping process in the grouping processing unit 31 will be specifically described. An image 81 illustrated in FIG. 2 is an image obtained by capturing the traveling direction of the host vehicle on which the object detection device 1 is mounted. For the image 81 in FIG. 2, the distance information acquisition unit 30 calculates the parallax as depth distance information for each pixel region. The parallax of the image 81 is calculated using, for example, the image on the other side of the left and right images captured simultaneously with the image 81. Then, the grouping processing unit 31 groups adjacent pixel areas within a predetermined range in the vertical direction of the image as one group. As the predetermined range, a parallax range set in the grouping processing unit 31 in advance may be used.

  In the grouping process, grouping may be performed using a parallax change state with respect to a vertical position. An image 82 in FIG. 3A is a schematic diagram of grouped images. In the image 82, adjacent or adjacent pixels in the same depth distance range in the vertical direction are set as the same group. For example, a plurality of groups such as groups 83a, 83b, and 83c are set and grouped.

  FIG. 3B is a graph showing the parallax with respect to the vertical position of the image in bb in FIG. As shown in FIG. 3B, three groups 83a, 83b, and 83c are set based on the parallax of the pixels. The group 83a is a group made up of pixels in the same depth distance range with a small parallax, and is recognized as representing a distant background due to the small parallax. The group 83b is a group composed of pixels in the same depth distance range, and since the parallax is not small, it is recognized as a three-dimensional object or an object to be detected, not a background. The group 83c is a group in which pixels in the same depth distance range are adjacent to each other, and changes so that the parallax increases as it goes to a lower position. For this reason, the group 83c is recognized as representing a road surface. The grouping processing unit 31 records and stores the grouped group result. For this reason, the grouping processing unit 31 also functions as a group recording unit. The group result is recorded each time grouping is performed. The recorded group result can be used to detect the movement of the object represented by the group.

  In FIG. 1, the movement information acquisition unit 32 acquires an optical flow including information on the movement direction and movement amount of the imaging target on the image for each pixel region of the image. For example, the movement information acquisition unit 32 calculates an optical flow for each pixel area of the image captured by the camera 2 and acquires the optical flow as movement information. The optical flow is a motion vector of a pixel area, and represents an object to be imaged on an image as a vector. For this reason, the optical flow has information on the moving direction and moving amount of the imaging target displayed in the pixel area. As described above, the pixel region may be a region composed of one pixel or a region composed of a plurality of pixels. When the pixel area is set as an area composed of a plurality of pixels, for example, an average value, a maximum value, a minimum value, or a predetermined representative value of the optical flows in the plurality of pixels may be used as the optical flow of the pixel area. The pixel area for acquiring the optical flow is set as an area having the same size as the pixel area for acquiring the depth distance information, but may be set as an area having a size different from that of the pixel area for acquiring the depth distance information.

  The specific calculation of the optical flow is performed by analyzing the movement of the pixel region using the image captured earlier with respect to the current image, and may be performed using a matching method or a gradient method. An image 84 in FIG. 4 schematically represents the optical flow 85. The optical flow 85 may be calculated excluding the area representing the background. For example, as shown in FIG. 4, the calculation of the optical flow 85 in the upper area 86 of the preset image, that is, the acquisition of the movement information may be omitted. By omitting this processing, the processing load for object detection can be reduced.

  The associating unit 33 associates the optical flows acquired by the movement information acquiring unit 32 with the pixel areas grouped by the grouping processing unit 31. For example, the optical flow acquired by the movement information acquisition unit 32 is associated with the pixel region set as a group by the grouping processing unit 31. At this time, the associating unit 33 associates optical flows by reducing the number of optical flow associations per unit area of an image for a group of pixel regions having a shorter depth distance. The number of optical flow associations may be calculated using a mathematical formula so that the depth distance decreases as the depth distance decreases, or is calculated using a map or table of optical flow associations corresponding to the depth distance. May be. At this time, the number of optical flow associations may be set so as to change continuously according to the depth distance, or may be set so as to change stepwise according to the depth distance. The specific association processing is performed, for example, by setting the optical flow data at the same coordinate position on the image as the optical flow of the pixel region for the pixel region set as a group.

  5 and 6 are explanatory diagrams of the association processing in the grouping processing unit 31. FIG. FIG. 5A shows an image 84 obtained by acquiring an optical flow 85 for each pixel region 84a. FIG. 5B shows an image 87 in which the optical flow 85 is associated with the grouped pixel region 87a. FIG. 6A shows an image 84 that has passed a predetermined time t from the image 84 in FIG. 5A and obtained an optical flow 85 for each pixel region 84a. FIG. 6B shows an image 87 in which a predetermined time t has elapsed from the image 87 in FIG. 5B, in which the optical flow 85 is associated with the grouped pixel region 87a. . A group 83d in the image 87 of FIGS. 5B and 6B is a group of image regions 87a representing vehicles. The vehicle in FIG. 5B is located far away, but in FIG. 6B, the vehicle is shown in close proximity as time elapses. The vehicle display in FIGS. 5B and 6B is an image display, and the ratio of the display size of the vehicle and the size of the pixel region 87a does not necessarily match the actual one.

  5B and 6B, the optical flow 85 is associated with the pixel region 87a of the group 83d. The group 83d in FIG. 6B is configured by a pixel region 87a that is closer in depth to the group 83d in FIG. 5B, and therefore the optical flow 85 is closely associated with each other. The number of associations of the optical flow 85 is reduced. In other words, the interval with which the optical flow 85 is associated (the skip interval) is larger as the group has a shorter depth distance. For example, when the same object approaches from a distance, the closer the object is, the larger the object is displayed on the image, but the optical flow 85 associated with the object does not increase accordingly.

  The associating unit 33 may associate the optical flows on the image 87 so that the optical flows are associated with the real space, which is the imaging space, at a preset interval. For example, the optical flow may be associated with the image 87 so that the optical flow 85 that is the movement information is associated with an interval of H (for example, 50 cm) in the real space. In this case, the interval h for associating the optical flow 85 can be set using, for example, an equation h = f · H / Z. In this equation, h is an association interval on an image, and f is a focal length converted to a pixel. H is the associating interval distance in the real space, and Z is the depth distance. By using this mathematical formula, the interval with which the optical flow 85 is associated becomes larger as the depth distance is closer, and the number of associations of the optical flow 85 per unit area is reduced.

  Further, instead of this mathematical formula, a map or table for setting the association interval according to the depth distance is used to calculate the association interval, that is, the number of associations of the optical flow 85 per unit area of the image. May be. For example, the optical flow 85 may be associated by setting an association interval whose depth distance is less than 5 m, an association interval of 5 m or more and less than 30 m, and an association interval of 30 m or more. The distance value of this association interval is an example, and may be other distance values. Further, the number of association intervals is not limited to three, but may be two or four or more. In this way, by setting the association interval to be larger as the depth distance is closer, the number of associations of the optical flow 85 per unit area of the image can be reduced as the depth distance is closer.

  Further, a threshold value for the depth distance is set, an association interval when the depth distance is less than the threshold value, and an association interval when the depth distance is greater than or equal to the threshold value are set, and the optical flow 85 is associated. May be performed. The setting of the threshold value is an example, and the number of threshold values is not limited to one and may be plural. In this way, by setting the association interval when the depth distance is less than the threshold value to be larger than the association interval when the depth distance is greater than or equal to the threshold value, the closer the depth distance, the more optical per unit area of the image The number of associations of the flow 85 can be reduced.

  The association unit 33 has been described with respect to a case where the number of optical flow associations per unit area of the image is reduced as the group of pixel regions with a shorter depth distance. When performing the optical flow association by reducing the number of optical flow associations per unit area, the association unit 33 associates the optical flow per unit area of the image with a group of pixel regions having a greater depth distance. The optical flow is associated by increasing the number of.

  In FIG. 1, the object detection unit 34 detects the movement state of objects around the vehicle using the optical flow associated by the association unit 33. For example, a group associated with an optical flow is recognized as an object around the vehicle, and the movement state of the group is detected based on the optical flow. Based on the movement state of the group, the relative velocity of the object in the three-dimensional space is calculated. The moving state of the object may be an absolute speed or an acceleration / deceleration in addition to the relative speed. Based on the detection result of the moving state of the object, it may be determined whether or not the object is an obstacle to traveling of the host vehicle, and the determination result may be notified to the driver of the host vehicle.

  The distance information acquisition unit 30, the grouping processing unit 31, the movement information acquisition unit 32, the association unit 33, and the object detection unit 34 described above may be configured by introducing software or programs that realize the respective functions into the ECU 3. Good. Further, some or all of them may be configured by individual electronic control units.

  An output unit 4 is electrically connected to the ECU 3. The output unit 4 operates based on the detection result of the object, and corresponds to, for example, a notification device for a driver, a control device that executes driving control, and the like.

  Next, the operation of the object detection apparatus 1 according to this embodiment will be described.

  FIG. 7 is a flowchart showing object detection processing in the object detection apparatus 1 according to the present embodiment. The object detection process is performed by the ECU 3, for example, and is repeatedly executed at a predetermined cycle.

  As shown in step S10 in FIG. 7 (hereinafter simply referred to as “S10”, the same applies to other steps S), first, an image reading process is performed. The image reading process is a process of reading image data of an image captured by the camera 2. When a stereo camera is used as the camera 2, a plurality of image data are read. And a process transfers to S12 and the acquisition process of distance information is performed. The distance information acquisition process is a process of acquiring depth distance information in an image captured by the camera 2. In other words, in this acquisition process, the depth distance information is acquired for each pixel area constituting the image using image data. The depth distance information acquired in this acquisition process is repeatedly acquired at a predetermined cycle, and is recorded every time it is acquired. As a specific processing content, for example, when a stereo camera is used as the camera 2, processing for generating parallax images for two left and right images is performed. A parallax value for each pixel area in the parallax image is acquired as depth distance information. Further, the depth distance for each pixel area may be calculated based on the parallax for each pixel area, and the value of the depth distance may be acquired as the depth distance information.

  Then, the process proceeds to S14, and a grouping process is performed. The grouping process is a process of grouping a plurality of pixel regions in association with an image from which the depth distance information has been acquired based on the depth distance information. In this grouping process, adjacent pixel regions in the three-dimensional space are set as the same group. For example, as illustrated in FIG. 3A, when pixel areas having the same depth distance range are close to each other in the vertical direction of the image, the pixel areas are set as the same group. The same depth distance range is a range including substantially the same depth distance within a predetermined depth distance range with respect to the same depth distance and the same depth distance. Further, the grouping process may be performed by grouping as the same group when pixel regions within the same distance range are close to each other in the vertical and horizontal directions of the image.

  Then, the process proceeds to S16, and a group reliability setting process is performed. The group reliability setting process is a process for setting the reliability of whether or not the grouped pixel region is for displaying a three-dimensional object. For example, it is determined whether the pixel areas grouped based on the depth distance information of the pixel area indicate a three-dimensional object. A three-dimensional object is not a background or a road surface but a three-dimensional object including an object to be detected. As shown in FIG. 3B, when the groups 83a, 83b, and 83c are set, based on the parallax value that is the depth distance information and the parallax change state according to the position on the image, the groups 83a, It is determined whether or not 83b and 83c each display a three-dimensional object. For example, if the parallax value is not a value indicating a distant distance greater than or equal to a predetermined value, and the change state of the parallax is not equal to or greater than the predetermined value according to the position on the image, it may be determined as a three-dimensional object. On the other hand, if the parallax value is a value indicating a distant distance greater than or equal to the predetermined value, or the change state of the parallax value is equal to or greater than the predetermined value according to the position on the image, it may be determined that the object is not a solid object. The group 83a has a parallax value change state that is not greater than or equal to a predetermined value depending on the position, but is a value that indicates a distant distance that is greater than or equal to a predetermined value, and is determined not to be a three-dimensional object but to indicate a background. In the group 83b, the disparity value is not a value indicating a distant distance greater than or equal to a predetermined value, and the change state of the disparity value is not equal to or greater than the predetermined value depending on the position. For this reason, it determines with it showing a solid thing. In the group 83c, the parallax value is not a value indicating a distant distance greater than or equal to a predetermined value, but the parallax value is inclined more than a predetermined value with respect to the position, and the change state of the parallax value is equal to or greater than the predetermined value depending on the position It has become. For this reason, it is determined not to indicate a three-dimensional object but to indicate a road surface. Thus, it is determined that the groups 83a and 83c have a group reliability not higher than a predetermined value and do not indicate a solid object, and the group 83b has a group reliability higher than a predetermined value and indicates a solid object. .

  And a process transfers to S18 of FIG. 7, and the acquisition process of movement information is performed. The movement information acquisition process is a process of acquiring an optical flow including the movement direction and movement amount of the imaging target on the image for each pixel region of the image. For example, in this acquisition process, an optical flow is calculated for each pixel area of the image captured by the camera 2 and acquired as movement information. As described above, a known method such as a matching method or a gradient method can be used for specific calculation of the optical flow.

  And a process transfers to S20 and the matching process of movement information is performed. The movement information association process is a process of associating the optical flows acquired in the movement information acquisition process in S18 with the pixel areas grouped in the grouping process in S14. This association process may be performed for all the groups set in the image, or may be performed only for the group determined as a group indicating a three-dimensional object in S16. As the grouping process, for example, optical flows at the same position on the image are associated with the pixel areas set as a group in the grouping process of S14. At that time, optical flow association is performed by reducing the number of optical flow associations per unit area of an image for a group of pixel regions having a shorter depth distance. As described above, the number of optical flow associations may be calculated using a mathematical formula so that the depth distance decreases as the depth distance decreases, or an optical flow association map corresponding to the depth distance or You may calculate using a table. At this time, the number of optical flow associations may be set so as to change continuously according to the depth distance, or may be set so as to change stepwise according to the depth distance. Furthermore, as the group reliability calculated in S16 is higher, the number of optical flow associations per unit area of the image may be increased to perform optical flow association. For example, a group or table that determines a group with a high group reliability for a group in which the change state of the parallax difference with respect to the position on the image is smaller than a predetermined threshold and associates the optical flow more densely may be used. A weight may be applied to the mathematical formula for calculating the number of associations.

  Then, the process proceeds to S22, and a group motion calculation process is performed. This calculation process is a process for calculating the movement of the grouped pixel regions, and is performed by the object detection unit 34. For example, the movement state of an object around the vehicle is detected using the optical flow associated by the association process of S20. That is, a group associated with an optical flow is recognized as an object, and the movement state of the group is detected based on the optical flow. Specifically, the relative velocity of the object in the three-dimensional space is calculated based on the moving state of the group. The moving state of the object may be an absolute speed or an acceleration / deceleration in addition to the relative speed. The optical flow of the group detected by this calculation process is recorded and saved and used in the calculation process executed in the next and subsequent cycles.

  And a process transfers to S24 and an output process is performed. The output process is a process for outputting an object detection result. For example, the detection result of the object is output to the output unit 4 such as a notification device that notifies the driver and a control device that executes the driving control. When the process of S24 is finished, a series of control processes are finished.

  In the series of control processes in FIG. 7, the order of the control processes may be changed or a part of the control processes may be omitted as long as the control result is not affected.

  As described above, according to the object detection apparatus 1 according to the present embodiment, in object detection using an image, the number of optical flows associated per unit area of an image is smaller in a group of pixel regions having a shorter depth distance. Thus, optical flow is associated. Thereby, the load of image processing is reduced as the group of pixel regions has a shorter depth distance. At this time, since the group of pixel areas with a close depth distance is displayed with a large object on the image, the detection accuracy of the object is not greatly reduced even if the number of optical flow associations is reduced. Accordingly, it is possible to reduce the load of the object detection process while suppressing a decrease in the object detection accuracy in the group of pixel regions having a short depth distance.

  For example, when object detection is performed on the vehicle imaged in the image 81 of FIG. 2, when the vehicle is far away, the vehicle is displayed small, and the number of pixel areas for displaying the vehicle is small. On the other hand, as the vehicle approaches, it is displayed larger on the image 81 and the number of pixel areas for displaying the vehicle increases. At this time, when the pixel areas for displaying the vehicle are set as the same group, when the optical flow is associated with all the pixel areas for displaying the vehicle regardless of the distance of the vehicle, when the vehicle comes close There are a large number of pixel areas for displaying a vehicle, and the process of associating optical flows becomes a heavy load. Moreover, substantially the same optical flow is associated with the pixel area displaying the vehicle, and the accuracy of object detection is not greatly improved. On the other hand, in the object detection apparatus 1 according to the present embodiment, the closer the object displayed, the smaller the number of optical flow associations, thereby suppressing the decrease in the object detection accuracy and the object detection processing. The load can be reduced. As a result, the object detection process can be performed even using an image processor with a small memory capacity. Further, the object detection can be speeded up by reducing the image processing.

  Further, in the object detection device 1 according to the present embodiment, the optical flow is associated with the pixel area so that the optical flows are associated with each other at a predetermined interval in the real space, whereby the object to be imaged is captured. The closer the depth distance is, the smaller the number of optical flow correspondences can be made. Therefore, it is possible to reduce the load of the object detection process while suppressing a decrease in detection accuracy of an object having a short depth distance.

  In addition, embodiment mentioned above demonstrates one Embodiment of the object detection apparatus which concerns on this invention, The object detection apparatus which concerns on this invention is not limited to what was described in the said embodiment. The object detection apparatus according to the present invention may be a modification of the object detection apparatus according to the above-described embodiment or application to other objects without changing the gist described in each claim.

  DESCRIPTION OF SYMBOLS 1 ... Object detection apparatus, 2 ... Camera, 3 ... ECU, 4 ... Output part, 30 ... Distance information acquisition part, 31 ... Grouping process part, 32 ... Movement information acquisition part, 33 ... Association part, 34 ... Object detection Part, 85 ... Optical flow.

Claims (2)

In an object detection device that detects an object around the vehicle using an image captured by a camera mounted on the vehicle,
A distance information acquisition unit that acquires depth distance information including a depth distance in the image, for each pixel region constituting the image;
A grouping processing unit that groups a plurality of the pixel regions based on the depth distance information;
A movement information acquisition unit that acquires an optical flow including information on a movement direction and a movement amount of the imaging target on the image for each pixel region with respect to the image;
An associating unit for associating the optical flows acquired by the movement information acquiring unit with the pixel regions grouped by the grouping processing unit;
An object detection unit that detects a moving state of an object around the vehicle using the optical flow associated by the association unit;
The association unit associates the optical flows by reducing the number of associations of the optical flows per unit area of the image as the group of the pixel regions having a shorter depth distance.
Object detection device.   The object detection device according to claim 1, wherein the association unit associates the optical flow with the pixel region so that the optical flow is associated with a predetermined interval in real space.

Images