JP2007088601A - Object detection device - Google Patents

Abstract

An object detection apparatus capable of appropriately detecting a part of a hidden detection target is provided.
An object detection apparatus that detects a detection target by performing image processing on a captured image, detects a vehicle in the image, performs edge processing on the image, and extends from an end of the vehicle. A lower end edge line is detected, and a vehicle in which a part of the vehicle is hidden is detected based on the lower end edge line. Thereby, vehicle detection can be performed according to the hidden ratio of the vehicle, and the detection accuracy of the vehicle can be improved.
[Selection] Figure 1

Description

  The present invention relates to an object detection apparatus that detects a detection target by performing image processing on a captured image.

Conventionally, as a device for detecting a detection target by performing image processing on a captured image, as described in JP-A-8-329397, a device for detecting a vehicle as a detection target, What differentiates image data, performs horizontal edge processing and vertical edge processing on the differential image, generates an image in which the horizontal edge amount and vertical edge amount are emphasized, and recognizes the preceding vehicle based on the image Are known.
JP-A-8-329397

  However, in such an apparatus, it is difficult to detect a plurality of detection objects that are imaged in an overlapping manner. For example, when a vehicle traveling in front of a preceding vehicle is partially hidden behind the preceding vehicle, the entire vehicle is imaged for the partially hidden vehicle. It is difficult to recognize.

  Therefore, an object of the present invention is to provide an object detection apparatus that can appropriately detect a partially hidden detection target by performing image processing on a captured image.

  That is, the object detection device according to the present invention is an object detection device that detects a detection target by performing image processing on a captured image, and a first detection unit that detects a first detection target in the image; Second detection for detecting a second detection target partly hidden by the first detection target based on an edge processing means for edge processing the image and an edge line extending from an end of the first detection target And means.

  According to the present invention, it is possible to appropriately detect the second detection target that is partially hidden behind the first detection target by detecting the edge line extending from the end of the first detection target.

  Moreover, in the object detection apparatus according to the present invention, the second detection means calculates a hidden ratio of the second detection target based on the edge line, and detects the second detection target according to the hidden ratio. Preferably it is done.

  Further, in the object detection apparatus according to the present invention, the second detection means corrects the detection determination model according to the hidden ratio of the second detection target, and uses the corrected detection determination model to detect the second detection target. It is preferable to detect an object.

  According to these inventions, by detecting the second detection object in accordance with the hidden ratio of the second detection object, the second detection object can be handled even when the second detection object has a different hidden ratio. Objects can be detected accurately. For this reason, the detection accuracy of the second detection object can be improved.

  Moreover, the object detection apparatus according to the present invention is an apparatus that detects a first vehicle as the first detection target and detects a second vehicle partially hidden behind the first vehicle as the second detection target. And said 2nd detection means detects said 2nd vehicle based on the lower end edge line of said 2nd vehicle extended from the edge part of said 1st vehicle, It is characterized by the above-mentioned.

  According to this invention, it is possible to appropriately detect the second vehicle partially hidden by the first vehicle by detecting the lower edge line of the second vehicle extending from the end of the first vehicle.

  In the object detection device according to the present invention, the second detection means may determine a hidden ratio of the second vehicle based on the lower edge line and detect the second vehicle according to the hidden ratio. preferable.

  Further, in the object detection apparatus according to the present invention, the second detection means corrects the detection determination model according to the hidden ratio of the second vehicle, and detects the second vehicle using the corrected detection determination model. Preferably it is done.

  According to these inventions, by detecting the second vehicle according to the hidden ratio of the second vehicle, it is possible to accurately detect the second vehicle even when the second vehicle has a different hidden ratio. For this reason, the detection accuracy of the second vehicle can be improved.

  According to the present invention, it is possible to appropriately detect a detection target partially hidden by performing image processing on a captured image.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the description of the drawings, 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 according to an embodiment of the present invention.

  As shown in FIG. 1, an object detection apparatus 1 according to the present embodiment is an apparatus that detects a detection target by performing image processing on a captured image, and detects a vehicle as the detection target. The object detection device 1 is mounted on a vehicle, for example, and detects other vehicles that travel in front of or behind the object detection device 1. The object detection apparatus 1 includes an imaging unit 2, a vehicle detection unit 3, a model generation unit 4, a vehicle position estimation unit 5, a model correction unit 6, a vehicle determination unit 7, and an output unit 8. The vehicle detection unit 3, the model generation unit 4, the vehicle position estimation unit 5, the model correction unit 6, and the vehicle determination unit 7 are configured by, for example, an image processing IC and an image memory.

  The imaging unit 2 functions as an imaging unit that images a vehicle to be detected, and includes, for example, an imaging device and an image storage memory. An imaging device is installed so that the front or back of a vehicle can be imaged. Moreover, as an imaging device, a CCD camera, a C-MOS camera, etc. are used, for example. An image captured by the imaging device is recorded in an image storage memory. For example, as illustrated in FIG. 2, in the image 10 to be captured, the following vehicles 21, 22, and 23 are captured. The vehicle 21 is a vehicle that travels immediately behind the host vehicle, and the entire vehicle is imaged. The vehicle 22 is a vehicle that travels following the vehicle 21, a part of which is hidden behind the vehicle 21, and a part of the vehicle is imaged. The vehicle 23 is a vehicle that travels following the vehicle 22, a part of which is hidden behind the vehicle 22, and a part of the vehicle is imaged.

  In FIG. 1, a vehicle detection unit 3 detects a vehicle in which the entire vehicle is captured in a captured image by image processing. The vehicle detection unit 3 has a vehicle detection memory, and records the detected position and size of the vehicle in the vehicle detection memory.

  The model generation unit 4 generates a model used for vehicle detection. The model generation unit 4 includes a database storage memory and a model storage memory. The database storage memory is a memory that stores a vehicle sample image in which the vehicle is imaged, a background image in which the vehicle is not imaged, and the like. For example, as shown in FIG. 3, the vehicle sample image 31 is an image obtained by cutting out the vehicle 31a. The vehicle 31a is cut out with a width a on the lower side, a width b on the left side, and a width c on the right side when the vehicle sample image 31 has a size of height h and width w.

  The model generated by the model generation unit 4 is composed of a plurality of filters for detecting a portion representing the vehicle. For example, as illustrated in FIG. 4, the model 40 includes a plurality of filters 41 that are arranged at an outer edge portion or a boundary portion of the vehicle 31 a. The filter 41 may be, for example, a boundary between a vehicle bumper and a shaded portion under the vehicle, a boundary between a vehicle roof and a background, a boundary between a windshield and a hood, a boundary between a tire and a road background, a boundary between a vehicle and a lateral edge. Placed in. The relative positions of the filters 41 are set so that their relative positional relationship does not change even if the size of the model 40 is increased or decreased. Thereby, vehicle detection can be performed by changing the size of the model 40 with respect to vehicles of different sizes.

  In FIG. 4, only the region where the filter 41 is arranged is shown. For example, as shown in FIG. 5, the filter 41 is configured by providing a positive region 41 a and a negative region 41 b with the outer edge or boundary of the vehicle as a boundary. The filter value is calculated by subtracting the sum of the luminance values of the pixels in the negative region 41b from the sum of the luminance values of the pixels in the positive region 41a. Based on the magnitude of this filter value, the light / dark difference between the positive region 41a and the negative region 41b can be determined, and a partial determination of whether or not the vehicle is a vehicle can be made.

  In this model generation part 4, it is preferable to generate a plurality of models based on a plurality of vehicle sample images. As a result, models corresponding to various vehicles can be generated, and the detection accuracy of various vehicles can be improved. For each filter constituting the model, it is preferable to set a weight according to the detection rate and false detection rate of vehicle detection. Thereby, the detection accuracy of the vehicle is improved.

  The vehicle position estimation unit 5 functions as vehicle position estimation means for estimating the position of the vehicle in the image by image processing. The vehicle position estimation unit 5 estimates the position of the vehicle, for example, by performing edge processing of the captured image, extracting the end point of the vehicle based on the edge, and detecting the tire.

  The model correction unit 6 corrects a model used for vehicle detection. That is, the hiding degree of the vehicle is detected, and the model for detecting the entire vehicle is corrected according to the hiding degree. The vehicle determination unit 7 determines whether there is a vehicle in the image. The vehicle determination unit 7 can determine whether a partially hidden vehicle is a vehicle. The output unit 8 is an output unit that outputs a detection result of the vehicle. For example, a unit that outputs a visual signal such as a lamp, a liquid crystal, or an LED, or a unit that outputs a visual signal such as a buzzer or a speaker is used.

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

  In FIG. 1, imaging of a vehicle traveling behind the host vehicle is performed by the imaging unit 2. The captured image captured by the imaging unit 2 is input to the vehicle detection unit 3 and subjected to image processing for vehicle detection.

  FIG. 6 is a flowchart of the vehicle detection process in the object detection apparatus 1. The series of control processes in FIG. 6 are repeatedly executed at a predetermined cycle with the vehicle detection unit 3 as a main body.

  First, as shown in S10 of FIG. 6, a captured image reading process is performed. The captured image reading process is performed by taking the image captured by the imaging unit 2 into the vehicle detection unit 3. Then, the process proceeds to S12, and detection area setting processing is performed. The detection area setting process is a process of setting a detection area for detecting a vehicle in the captured image. For example, as illustrated in FIG. 7, a rectangular detection region 50 is set in the captured image 10. The size of the detection area 50 is set according to the size of the vehicle to be detected.

  And it transfers to S14 and the calculation process of a detection score is performed. The detection score calculation process is a process of calculating a vehicle detection score in the detection area 50. The detection score is a numerical value used for vehicle detection determination in the detection region 50. The detection score is calculated using a vehicle detection model 40 shown in FIG. That is, the size of the model 40 is adjusted to the detection area 50, and each filter value in the filter 41 of the model 40 is calculated. Then, the total sum of the filter values multiplied by the weights is calculated as the detection score. For example, the detection score is calculated as a larger value when a vehicle is present in the detection area 50 than when no vehicle is present.

  And it transfers to S16 of FIG. 6 and it is judged whether the detection score exceeded the vehicle detection threshold value. The threshold value is a set value set in advance in the vehicle detection unit 3. If the detection score does not exceed the threshold value in S16, it is determined that no vehicle is detected in the detection area 50, and the process proceeds to S20. On the other hand, if the detection score exceeds the threshold value in S16, it is determined that a vehicle is detected in the detection area 50, and a vehicle detection recording process is performed (S18). The vehicle detection recording process is a process of recording the position and size of the vehicle in the vehicle detection memory.

  Then, the process proceeds to S20, and the detection area 50 is shifted. This shift process is a process of moving the detection area 50 to the next detection position. For example, as shown in FIG. 7, the detection area 50 is moved rightward from the upper left position of the captured image 10, moved to the lower left edge when it reaches the right edge, and sequentially shifted until it reaches the lower right position. Is done.

  Then, the process proceeds to S22 to determine whether or not all the shifts are completed. That is, it is determined whether or not the detection area 50 has moved to the lower right position and all the detection positions have been moved. If it is determined that all the shifts have not been completed, the process returns to S14. On the other hand, when it is determined that all the shifts have been completed, a detection area size changing process is performed. The size changing process is a process for changing the size of the detection area 50. For example, the detection area 50 can be set from the minimum size to the maximum size, and is changed to a plurality of sizes from the minimum size to the maximum size.

  Then, the process proceeds to S26, and it is determined whether or not all the size changes are completed. That is, it is determined whether or not the detection area 50 has been resized from the minimum size to the maximum size. If it is determined that all the size changes have not been completed, the process returns to S14. On the other hand, when it is determined that all the size changes have been completed, a series of control processes of the vehicle detection process is ended.

  According to such a vehicle detection process, the vehicle 21 that appears entirely in the captured image 10 can be detected, and the position and size of the vehicle 21 on the captured image 10 can be detected. For example, in FIG. 2, the position and size of the vehicle 21 that appears as a whole can be detected.

  In this vehicle detection process, it is preferable to use a plurality of vehicle detection models. For example, various vehicles can be accurately detected by performing vehicle detection using a plurality of models corresponding to different types of vehicles.

  Next, the vehicle position estimation process in the object detection apparatus 1 will be described.

  FIG. 8 shows a flowchart of the vehicle position estimation process. This vehicle position estimation process is a process of estimating the position of a partially hidden vehicle based on the position information of the entire vehicle detected by the vehicle detection process described above.

  In this vehicle position estimation process, edge processing is first performed as shown in S30. The edge processing is processing for generating an edge image by performing edge processing on the captured image 10. And it transfers to S32 and the detection process of the lower end edge of the vehicle 21 is performed. This detection process is a process for detecting the lower edge of the vehicle 21 that appears entirely based on the edge image.

  For example, as shown in FIG. 9, a window 12 for detecting the vehicle lower edge is set in the edge image 11, and the window 12 is shifted in the horizontal direction. And as shown in FIG. 10, the intersection of the window 12 and a horizontal edge is graphed. The window 12 is set so as to include at least the vehicle 21 by shifting based on the position and size of the vehicle 21 detected in the vehicle detection process.

  In FIG. 10, the horizontal edge line 61 that is longer than a predetermined length and appears at the bottom is detected as the lower edge of the vehicle 21. Usually, since the space between the lower end of the vehicle and the road surface becomes dark due to a shadow, a horizontal edge is formed at a position below the vehicle. For this reason, the lower end edge of the vehicle can be accurately detected by extracting a horizontal edge line having a length equal to or longer than a predetermined length in the edge image.

  Then, the process proceeds to S34 in FIG. 8, and the detection processing of the lower edge of the subsequent vehicle is performed. This detection process is a process for detecting the lower edge of a vehicle that is partially hidden. For example, in FIG. 10, as the window shifts, the lower edge 61 of the vehicle 21 disappears and the lowermost part moves to the next edge 62. If the difference Y1 in the vertical position between the edge 62 and the lower edge 61 is within a predetermined value and the edge 62 is longer than a predetermined length, it is determined that the edge 62 is a lower edge of a hidden vehicle. The

  Then, the process proceeds to S36 in FIG. 8 to determine whether or not there is a lower end edge of the following vehicle. When it is determined that there is no lower edge of the following vehicle, the process proceeds to S48. On the other hand, when it is determined in S36 that there is a lower edge of the following vehicle, a vertical edge detection process is performed (S38). The vertical edge detection process is a process for detecting a vertical edge formed between vehicles. For example, as shown in FIG. 11, a vertically long window 13 is set at a position between the vehicle 21 and the vehicle 22, and it is detected whether or not there is a vertical edge having a predetermined length or longer in the setting region of the window 13. The position between the vehicle 21 and the vehicle 22 may be set to a position corresponding to the boundary position P1 between the lower edge 61 and the lower edge 62 in FIG.

  Then, the process proceeds to S40 in FIG. 8, and it is determined whether or not a vertical edge is detected. If no vertical edge is detected, the process proceeds to S48. On the other hand, if a vertical edge is detected in S40, tire detection processing is performed (S42). The tire detection process is a process for detecting a tire of a vehicle, and is performed by specifying the tire position of the vehicle from the position of the vehicle and detecting whether there is a tire at the tire position. For example, as shown in FIG. 12, tire detection filters 41 and 41 are applied to the tire position of the captured image 10, and the presence or absence of the tire is evaluated based on the brightness value.

  Then, the process proceeds to S44 in FIG. 8, and it is determined whether or not a tire is detected. If no tire is detected, the process proceeds to S48. On the other hand, when a tire is detected in S44, a recording process is performed (S46). The recording process is a process of recording the lower end position, the upper end position, the lower corner position, and the size of the vehicle. The lower end position of the vehicle is recorded based on the position of the lower end edge. The upper end position of the vehicle is recorded on the basis of an upper end edge having a predetermined length or more and located above the lower end edge. The lower corner position of the vehicle is recorded based on the vehicle tire position. The size of the vehicle is calculated and recorded based on the lower edge of the vehicle.

  When the recording process is completed, the process returns to the lower edge detection process of the succeeding vehicle in S34. In this case, in the second lower edge detection process, the lower edge of the vehicle following the succeeding vehicle is detected. That is, as shown in FIG. 2, when there is a vehicle 21 that appears as a whole, and there are a vehicle 22 and a vehicle 23 following the vehicle 21, a lower edge detection process of the vehicle 23 following the vehicle 22 is performed. . Then, the vertical edge detection (S38), the tire detection (S42), and the recording process (S46) are sequentially performed, whereby the lower end position, the upper end position, and the lower corner position of the vehicle 22 and the vehicle 23 are recorded.

  By the way, when it is determined in S36 that there is no lower edge of the following vehicle, the process proceeds to S48, and the recording and storing process of the vehicle end position is performed. In this case, it is determined that there is no subsequent vehicle, and the end position of the detected vehicle is recorded and stored. For example, in FIG. 2, the vehicle lower left positions 21 a, 22 a, and 23 a are recorded for the vehicles 21, 22, and 23.

  Then, the process proceeds to S50, and it is determined whether or not there is a vehicle in which another whole appears. When there is a vehicle in which another whole appears in the captured image 10, the process proceeds to S32. On the other hand, if there is no other vehicle in which the whole appears in S50, the series of control processing of the vehicle position estimation processing is ended.

  According to such a vehicle position estimation process, it is possible to appropriately estimate the positions of the entire vehicle (the vehicle in which the entire image appears) and the partial vehicle (the vehicle in which the image is partially hidden) captured in the captured image 10. it can. That is, by detecting the lower edge of the vehicle and detecting the vertical edge between the vehicles, some vehicles can be detected appropriately.

  In addition, by detecting the tire when detecting the position of the vehicle, the certainty of the presence of the vehicle is improved, and the accuracy of detecting the position of the vehicle can be improved.

  Next, the vehicle determination process in the object detection apparatus 1 will be described.

  FIG. 13 shows a flowchart of the vehicle determination process. This vehicle determination process is a process for detecting a partially hidden vehicle.

  In this vehicle determination process, as shown in S60, a model correction process is first performed. The model correction process is a process of correcting a model for detecting a vehicle that has appeared entirely to a model for detecting a partially hidden vehicle. For example, as shown in FIG. 14, in the captured image 10, a distance d1 from the lower left position 22a of the vehicle 22 to the vehicle 21 is calculated. This distance d1 indicates the hidden ratio of the vehicle 22. Based on the distance d1 and the vehicle size, the vehicle detection model is corrected. This model correction is performed based on the distance d1 and the vehicle size, so that the model can be corrected according to the hidden ratio of the vehicle 22. As the vehicle size, the one calculated in the vehicle position estimation process is used.

  As shown in FIG. 15, the model correction is performed by deleting a part of the vehicle detection model 40 as much as the vehicle 22 is hidden. The deletion range 40a is set according to the hiding ratio based on the size of the vehicle 22 and the appearing distance d1. Further, the vehicle detection threshold value is corrected in accordance with the partial deletion of the model 40. Since the detection score value is decreased by the amount of the filter 41 to be deleted and the filter 41 to be partially deleted, the vehicle detection threshold value is decreased in accordance with the decrease. As a result, it is possible to make an appropriate vehicle detection determination of the vehicle 22 in which only a part appears.

  Then, the process proceeds to S62 in FIG. 13 and detection score calculation processing is performed. The detection score calculation process is a process of calculating the detection score of the vehicle 22 using the model corrected in S60. For example, with respect to the captured image 10, the correction model is arranged at the position of the vehicle 22, the filter value of each filter 41 of the correction model is calculated, and the detection score that is the sum of the filter values is calculated.

  And it transfers to S64 and it is judged whether the detection score exceeded the vehicle detection threshold value. If the detection score does not exceed the vehicle detection threshold, it is determined that no vehicle has been detected, and the process proceeds to S68. On the other hand, when the detection score exceeds the vehicle detection threshold value in S64, the vehicle detection recording process is performed assuming that the vehicle is detected (S66). The vehicle detection recording process is a process of recording information about the vehicle 22 in which only a part appears. For example, the position and size of the vehicle 22 are recorded.

  Then, the process proceeds to S68, where it is determined whether there is another vehicle to be determined. For example, as shown in FIG. 14, when there is a vehicle 23 following the vehicle 22, it is determined that there is another vehicle to be determined, and the process returns to S <b> 60 and model correction corresponding to the vehicle 23 is performed. On the other hand, when it is determined in S68 that there is no other vehicle to be determined, a series of control processes of the vehicle determination process is ended.

  According to such a vehicle determination process, vehicle detection can be performed appropriately even for a partially hidden vehicle. That is, for a partially hidden vehicle, model correction is performed according to the degree of hiding, and appropriate vehicle detection is possible by using the corrected model.

  As described above, according to the object detection device 1 according to the present embodiment, a vehicle whose part is hidden by the vehicle is detected by detecting the lower end edge line of the vehicle extending from the end portion of the vehicle where the whole appears. It can be detected properly.

  In addition, by detecting the vehicle according to the hidden ratio of the vehicle, the vehicle can be accurately detected even when the vehicle has a different hidden ratio. For this reason, it is possible to improve the detection accuracy of a partially hidden vehicle.

  The above-described embodiment shows an example of the object detection apparatus according to the present invention. The object detection device according to the present invention is not limited to this, and the object detection device according to the embodiment may be modified or otherwise provided that the gist described in each claim is not changed. It may be applied to. For example, in the present embodiment, an apparatus for detecting a vehicle has been described. However, the present invention may be applied to detection of an object other than a vehicle.

1 is a schematic configuration diagram of an object detection device according to an embodiment of the present invention. It is explanatory drawing of the captured image in the object detection apparatus of FIG. It is explanatory drawing of the vehicle sample image recorded on the object detection apparatus of FIG. It is explanatory drawing of the model for vehicle detection used with the object detection apparatus of FIG. It is explanatory drawing of the model for vehicle detection used with the object detection apparatus of FIG. It is a flowchart of the vehicle detection process in the object detection apparatus of FIG. It is explanatory drawing of the vehicle detection process in the object detection apparatus of FIG. It is a flowchart of the vehicle position estimation process in the object detection apparatus of FIG. It is explanatory drawing of the vehicle position estimation process in the object detection apparatus of FIG. It is explanatory drawing of the vehicle position estimation process in the object detection apparatus of FIG. It is explanatory drawing of the vehicle position estimation process in the object detection apparatus of FIG. It is explanatory drawing of the vehicle position estimation process in the object detection apparatus of FIG. It is a flowchart of the vehicle determination process in the object detection apparatus of FIG. It is explanatory drawing of the vehicle determination process in the object detection apparatus of FIG. It is explanatory drawing of the vehicle determination process in the object detection apparatus of FIG.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Object detection apparatus, 2 ... Imaging part, 3 ... Vehicle detection part, 4 ... Model production | generation part, 5 ... Vehicle position estimation part, 6 ... Model correction part, 7 ... Vehicle determination part, 8 ... Output part.

Claims (6)

In an object detection apparatus that detects a detection target by performing image processing on a captured image,
First detection means for detecting a first detection object in the image;
Edge processing means for edge processing the image;
Second detection means for detecting a second detection object partially hidden behind the first detection object based on an edge line extending from an end of the first detection object;
An object detection apparatus comprising: The second detection means calculates a hidden ratio of the second detection object based on the edge line, and detects the second detection object according to the hidden ratio;
The object detection apparatus according to claim 1. The second detection means corrects the detection determination model according to the hidden ratio of the second detection target, and detects the second detection target using the corrected detection determination model.
The object detection device according to claim 1, wherein: A device that detects a first vehicle as the first detection object and detects a second vehicle that is partially hidden behind the first vehicle as the second detection object,
The second detection means detects the second vehicle based on a lower edge line of the second vehicle extending from an end of the first vehicle;
The object detection apparatus according to claim 1. The second detection means calculates a hidden ratio of the second vehicle based on the lower edge line, and detects the second vehicle according to the hidden ratio;
The object detection apparatus according to claim 4. The second detection means corrects the detection determination model according to the hidden ratio of the second vehicle, and detects the second vehicle using the corrected detection determination model;
The object detection apparatus according to claim 4, wherein:

Images