JP2013211005A - Image processing device and vehicle control system using the same - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image processing device configured to properly detect an object in an image.SOLUTION: A plurality of part models are stored for detecting parts as portions showing characteristics of an object from an image. An image from an imaging apparatus is input (S110). The part models stored in a part model storing part are used to calculate a score map, and a high-score area is extracted (S120). A part group area is extracted, accordingly (S130). On the basis of the part group area, a part area for each part is extracted (S140). A score of the extracted part area is calculated (S150), to estimate an object area where the object exists (S160).

Description

  The present invention relates to a technique for detecting a pedestrian from an image of a vehicle-mounted camera.

  Conventionally, a technique for detecting a pedestrian from an image of a vehicle-mounted camera has been proposed for the purpose of avoiding a collision with an object that becomes an obstacle to vehicle travel. For example, a technique for detecting a pedestrian by detecting an object to be discriminated from an image of an in-vehicle camera and applying a neural network object discrimination method using a pedestrian's contour pattern to the object has been proposed. Yes.

  However, in the neural network object discrimination method using the pedestrian contour pattern, an object such as a tree similar to the pedestrian contour pattern may be detected as a pedestrian, and the detection accuracy of the pedestrian decreases. May end up.

  As a technique for solving this problem, a technique has been proposed in which a determination regarding a predetermined part constituting a pedestrian is performed and a pedestrian is determined based on whether or not the sum of determination output values of each predetermined part exceeds a threshold value. (For example, refer to Patent Document 1). For example, a determination regarding a predetermined part such as a head, a torso, a hand, and a leg is performed, and detection is performed based on the sum of determination output values of each predetermined part.

JP 2008-21034 A

  However, the technique described in Patent Document 1 has a problem that a pedestrian behind a pedestrian or a pedestrian hidden behind a pedestrian cannot be accurately detected. That is, when some of the predetermined parts such as the pedestrian's head, torso, hand, and foot are not shown in the image, the determination output value of the predetermined part is reduced, so that the sum of the determination output values is reduced. Therefore, when a pedestrian is detected based on the sum of the determination output values, there is a concern that no pedestrian is detected.

Note that the problems listed here are not limited to pedestrians, but also occur when an object in an image is detected.
The present invention has been made to solve the above-described problems, and an object of the present invention is to provide an image processing apparatus capable of appropriately detecting an object in an image and vehicle control using the image processing apparatus. To provide a system.

The image processing device (10) according to claim 1 made to achieve the above object detects an object based on an image from an imaging device (32) mounted on a vehicle.
In this image processing apparatus, the part model storage means (61) stores a plurality of part models capable of detecting parts, which are parts indicating the characteristics of the object, from the image. For example, if the object is a pedestrian, the part model is a head part model corresponding to the head, an arm part model corresponding to the arm, a waist part model corresponding to the waist, and a leg part corresponding to the leg. Prepared as a model. For example, if the object is a vehicle, a tire part model corresponding to a tire, a taillight part model corresponding to a taillight, a window part model corresponding to a window, and the like are prepared.

  At this time, the part area extraction means (51) extracts the part area for each part using the part model stored in the part model storage means. Then, the object area estimation means (51) estimates the object area where the object exists based on the parts area extracted by the parts area extraction means.

  That is, the part area for each part is extracted using the part model, and the object area is estimated based on the part area. For example, even when only a part area corresponding to the leg part of the pedestrian is extracted, an area in which the pedestrian is an object is estimated from the part area corresponding to the leg part. Therefore, the object in the image can be detected appropriately.

It is a block diagram which shows schematic structure of a vehicle control system. It is a flowchart which shows a pedestrian detection process. It is explanatory drawing which shows extraction of the high score area using a parts model. It is explanatory drawing which shows the determination of the setting threshold value with respect to the score in a score map. It is explanatory drawing which shows extraction of a part group area | region. It is explanatory drawing which shows extraction of the part area | region based on a part group area | region, and calculation of the score of a part area | region. It is explanatory drawing which shows extraction of the pedestrian area | region from a parts area | region.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
The vehicle control system of the present embodiment is configured with the image processing device 10 and the vehicle control device 20 as the center. A distance measuring device 31 and an imaging device 32 are connected to the image processing device 10. In addition, a speaker 41, a brake 42 and a steering 43 are connected to the vehicle control device 20.

  The image processing device 10 and the vehicle control device 20 are configured as a so-called computer. As a result, the image processing device 10 detects a pedestrian based on the image from the imaging device 32, and determines the collision risk based on the distance information from the distance measuring device 31. On the other hand, the vehicle control device 20 performs notification by the speaker 41 according to the collision risk output from the image processing device 10, or performs avoidance control by the brake 42 and the steering 43 when the degree of emergency is large. To go.

Next, the configuration of each unit will be specifically described.
The distance measuring device 31 is embodied as a laser radar or the like. Of course, another configuration such as a millimeter wave radar may be used. The distance measuring device 31 can detect the distance to the object in front of the vehicle.

  The imaging device 32 is embodied as a monocular camera or the like. The imaging device 32 images the front of the vehicle. As described above, the image processing apparatus 10 detects a pedestrian based on this image.

  The speaker 41 is configured for audio output. When the risk of collision with a pedestrian increases, sounds such as “There is a pedestrian in front” and “Be careful” are output via the speaker 41.

  The brake 42 includes a brake driving device, and decelerates the vehicle by a signal from the vehicle control device 20. The steering 43 includes an EPS (Electric Power Steering) device, and assists the driver's steering by a signal from the vehicle control device 20. Such vehicle control is performed when the risk of collision with a pedestrian is further increased, and when the urgency is high.

  The image processing apparatus 10 includes a control unit 50 and a storage unit 60. The control unit 50 is embodied as a so-called microcomputer. The control unit 50 includes a pedestrian detection unit 51 and a collision risk determination unit 52 as functional blocks. The storage unit 60 is embodied as a ROM, for example. Of course, a flash memory or HDD other than the ROM may be used. The storage unit 60 includes a parts model storage unit 61 and a collision risk determination table storage unit 62.

  Here, the pedestrian detection unit 51 detects a pedestrian in the image output from the imaging device 32. Although specifically described later, each part of the pedestrian is detected by scanning an image, and the presence and position of the pedestrian are estimated.

  Since the distance to the pedestrian is known by the distance measuring device 31 after the pedestrian is detected, the collision risk level determination unit 52 determines the collision risk level according to the distance from the pedestrian. The collision risk is output to the vehicle control device 20.

  The part model storage unit 61 stores a part model. The part model is a model of each part of the pedestrian, and is used to detect the distribution of each part of the pedestrian in the image. In the present embodiment, the head, arms, waist, and legs are prepared as parts. Hereinafter, the part models corresponding to these parts are appropriately distinguished as “head part model”, “arm part model”, “waist part model”, and “leg part model”. The part model is a feature vector that expresses the feature of each part such as contour information.

  The collision risk determination table storage unit 62 stores a collision risk determination table. The collision risk determination table is a table that associates the distance to the pedestrian and the collision risk. The collision risk determination unit 52 determines the collision risk by referring to the collision risk determination table.

  The vehicle control device 20 includes a notification control unit 70 and a vehicle control unit 80. Both are embodied as so-called microcomputers. The notification control unit 70 is configured to perform notification via the speaker 41. When the collision risk level increases, first, notification via the speaker 41 is performed. The vehicle control unit 80 is configured to perform vehicle control via the brake 42 and the steering 43, and collision avoidance control is performed when the risk of collision further increases.

  In addition, this embodiment has the characteristic in the pedestrian detection process as a function of the pedestrian detection part 51. FIG. Therefore, the details of this pedestrian detection process will be described below. The pedestrian detection process is repeatedly executed at predetermined time intervals.

As shown in FIG. 2, image input is performed in the first S110. In this process, an image output from the imaging device 32 is input.
In subsequent S120, a score map is created and a high score area (hereinafter referred to as "high score area") is extracted. As described above, a part model obtained by modeling each part of a pedestrian is stored in the part model storage unit 61, but a score map is created by scanning an image using the part model. Further, a high score area having a score exceeding the set threshold is extracted.

  Specifically, as shown in FIG. 3A, the head part model TP1 is moved rightward at predetermined intervals, for example, sequentially from the upper left part of the image. When the right end of the image is reached, the image is returned to the left end and moved downward at a predetermined interval. Thereafter, the same movement is repeated, and the image is moved in order to the lower right part of the image. At this time, for each position of the head part model TP1, the similarity with the area overlapping the head part model TP1 is stored as a score. As a result, a score map (not shown) having a score for each part model position (each area) is created.

  In addition, a pedestrian present at a relatively far distance appears small in the image. Therefore, as shown in FIG. 3B, similar scoring is performed using a head part model TP2 obtained by reducing the head part model TP1.

  In addition, although the scanning in the two stages of the case where the pedestrian exists relatively near and the case where the pedestrian exists relatively far is described here, in reality, the scanning is performed in three or more stages. It is possible to scan. In the stepwise scanning, the part model may be enlarged / reduced, or the image to be scanned may be enlarged / reduced without changing the size of the part model.

  After creating such a score map, the position of the part model having a score exceeding the set threshold, that is, the high score area is extracted. As a result, as shown in FIG. 3C, high score areas TC and TT are extracted from the image. The high score area TC is an area where the score is high when scanned with the head part model TP1, and the high score area TT is an area where the score is high when scanned with the head part model TP2. is there. Since there are an arm part model, a lumbar part model, and a leg part model in addition to the head part model, a high score is obtained in the same manner as in FIG. An area is extracted.

  Here, an example of setting a threshold for determining a high score area will be described. In FIG. 4A, two persons J1 and J2 are shown in the image, but the arms and legs of the person J1 in the back are hidden by the arms and legs of the person J2 in the foreground. Therefore, paying attention to the fact that there are easily hidden parts in the pedestrian parts, as shown in FIG. 4 (b), the head K, the arms L1, L2, the waist M, and the legs N1, N2 In this embodiment, in the present embodiment, the threshold values for extracting the arm portions L1, L2 and the leg portions N1, N2 are set lower than the threshold values for extracting the hard-to-hide head K and waist M. Yes.

Returning to FIG. 2, in S130, a part group region is extracted. This process groups the high score areas extracted in S120.
Specifically, as shown in FIG. 5A, the high score areas TC and TT are grouped according to the mutual distance and the high score areas TC and TT. By extracting the pedestrians that exist in the vicinity and the pedestrians that exist in the distance using the size, the extraction accuracy is increased. FIG. 5A shows a state where a part group region G1 including the high score area TC and a part group region G2 including the high score area TT are extracted. The distance here is a 2D distance on an image or a 3D distance in real space (the same applies hereinafter).

  As shown in FIG. 5B, the group area may be extracted only by the mutual distance between the high score areas TC and TT. This is advantageous in that the processing becomes relatively simple.

  Further, the part group region may be extracted in consideration of the density of the high score area. This is because when the density is relatively low, it may be a false detection. Specifically, as shown in FIG. 5C, grouping is performed by the distance of the high score area TT, and part group regions G4 and G5 including the high score area TT are extracted. Then, the density is calculated by dividing the number of areas TT included in each part group region G4, G5 by the area of the part group region G4, G5. When the density exceeds a certain threshold value, the part group region is selected. When the density is equal to or less than the certain threshold value, the part group region is excluded.

  Furthermore, the part group region may be extracted in consideration of the total score in addition to or instead of the density. This is because if the total score is relatively low, it may be a false detection. Specifically, as shown in FIG. 5C, grouping is performed by the distance of the high score area TT, and part group regions G4 and G5 including the high score area TT are extracted. And the sum total of the score of the high score area TT contained in each part group area | region G4, G5 is calculated | required. Then, when the total score exceeds a certain threshold value, the part group region is selected, and when it is equal to or smaller than the certain threshold value, it is removed from the part group region.

  Returning to FIG. 2, in S140, a part region is extracted. In this process, a circumscribed rectangular area including the part group area extracted in S130 is extracted. At this time, if there is a part group area with a short distance, a circumscribed rectangular area including the group area is extracted. The distance between the part group regions may be the shortest distance from the boundary to the boundary, may be the distance of the center position of the part group region, or may be the distance of the center of gravity based on the score of each area. When such a distance is below a certain threshold, it can be considered that the distance is close. FIG. 6 shows extraction of a part region P that includes two part group regions G6 and G7.

In subsequent S150, the score of the part area is calculated. In this process, the score of the part area extracted in S140 is calculated.
Specifically, it is calculated as the sum of the scores of the part group areas included in the part area. The score of the part group area is calculated as the sum of the scores of the high score areas included in the part group area. For example, in FIG. 6, the scores of the part group regions G1 and G2 are calculated as the sum of the scores of the high score areas TT included in the part group regions G1 and G2, and the scores of the part region P are calculated as the sum of the scores of the part group regions G1 and G2. For example, the score is calculated. In this case, the score of the parts area can be calculated relatively easily.

  Further, when calculating the scores of the group regions G1 and G2, the scores may be calculated including the “positional proximity” of the high score area. Specifically, as shown in the lower part of FIG. 6, it is conceivable to define “closeness of position” of the high score area as the reciprocal of the difference sum of the positions p of the high score area.

  Furthermore, in addition to or instead of “closeness of position” of the high score area, the score of the part group area including “closeness of size” is assumed on the assumption that the high score area of different sizes is included in the part group area. May be calculated. Specifically, the “closeness of size” of each area can be defined as the reciprocal of the difference sum of the size s of the high score area.

  Further, in addition to or instead of these “position proximity” and “size proximity”, the score of the part group area may be calculated including the density of the part group area. The density in this case has already been described above.

  In subsequent S160, a pedestrian is detected. In this process, a pedestrian area, which is an area where a pedestrian exists, is extracted based on the extracted parts area. If the types of parts constituting the part area and the size of the high score area are known, the pedestrian area can be extracted.

  Specifically, as shown in FIG. 7A, when there is a part region P extracted based on the head part model, it is assumed that there is a body part below. Further, the size of the body part is estimated from the size of the high score area included in the part region P. Thereby, the pedestrian area S is estimated from the parts area P. In addition, when the score of the parts area | region P contained in the pedestrian area | region S is below a fixed threshold value, the pedestrian area | region S is excluded from extraction object.

  Also, specifically, as shown in FIG. 7B, there are two part areas P1 and P3 extracted based on the head part model, and there is a parts area P2 extracted based on the leg part model. In this case, the pedestrian area is detected based on the arrangement relationship. For example, when the high score areas included in the part areas P1 and P2 have the same size, the two part areas P1 and P2 are considered to have a correspondence relationship. That is, it can be considered as the head and legs of the same pedestrian group. On the other hand, if the size of the high score area included in the part area P3 is different from the size of the high score area included in the other part areas P1 and P2, and there is no part area corresponding to the part area P3, It is assumed that the region P3 is erroneous detection.

  And as shown in FIG.7 (c), when two part area | regions P1 and P2 have a correspondence, the circumscribed rectangle with respect to the pedestrian area | region S1, S2 estimated from the said part area | region P1, P2 is made into pedestrian area | region S3. Estimate as At this time, the pedestrian area S3 is estimated when the sum of the scores of the part areas P1 and P2 exceeds a certain threshold. When it is below a certain threshold value, the pedestrian area S3 is excluded from the extraction target.

  As described above in detail, according to the present embodiment, the part model storage unit 61 of the image processing apparatus 10 stores a plurality of part models that can detect parts that are parts of the pedestrian from the image. Has been. At this time, the pedestrian detection unit 51 of the control unit 50 inputs an image from the imaging device 32 (S110 in FIG. 2), and creates a score map using the part model stored in the part model storage unit 61. Then, a high score area is extracted (S120). Thereby, a part group area is extracted (S130), and a part area for each part is extracted based on the part group area (S140). And the score of the extracted parts area | region is calculated (S150), and the pedestrian area | region where a pedestrian exists is estimated (S160).

  That is, a part area for each part is extracted using a part model, and a pedestrian area is estimated based on the part area. For example, even when only the parts area corresponding to the leg part of the pedestrian is extracted, the pedestrian area is estimated from the part area corresponding to the leg part. Therefore, a pedestrian in the image can be detected appropriately.

  In the present embodiment, as shown in FIG. 3A, the head part model TP1 is moved in the image, and the degree of coincidence with the area that is an area overlapping the head part model TP1 is used as a score. A score map is created by calculation, and a high score area in which the score exceeds a preset threshold value is extracted (S120 in FIG. 2). That is, the part model is moved in the image, the degree of coincidence with the area that is an area overlapping with the part model is calculated as a score, and the part area is determined based on the high score area where the score exceeds a predetermined setting threshold. Extract. As a result, the parts region can be extracted relatively easily.

  At this time, in this embodiment, paying attention to the fact that there are parts that are also easily hidden in the pedestrian parts, as shown in FIG. 4B, the head K, arms L1, L2, waist M, Also, in each part of the leg parts N1 and N2, the threshold values for extracting the arm parts L1, L2 and the leg parts N1, N2 are set lower than the threshold values for extracting the head K and the waist part M which are difficult to hide. doing. That is, the setting threshold is determined for each part. Thereby, each part can be detected appropriately.

  In the present embodiment, the high score areas are grouped to extract a part group area (S130 in FIG. 2), and a parts area is extracted based on the parts group area (S140). As a result, a more appropriate part region can be extracted.

  Specifically, as shown in FIG. 5A, grouping is performed according to the distance between the high score areas TC and TT and the size of the high score areas TC and TT. That is, part group regions are extracted so as to include high score areas whose mutual distances are equal to or smaller than a certain threshold and have the same size. As described above, since the pedestrians present in the vicinity and the pedestrians present in the distance are separately extracted using the size, the extraction accuracy can be increased.

  Furthermore, in this embodiment, a part area is extracted as a circumscribed rectangular area that includes a part group area (S140 in FIG. 2). Extracting the part area as the circumscribed rectangular area is advantageous in that the subsequent processing becomes relatively simple.

  At this time, if there is a part group area with a short distance, a circumscribed rectangular area including the group area is extracted. For example, in the example of FIG. 6, a part region P that includes two part group regions G6 and G7 is extracted. That is, a part area is extracted as a circumscribed rectangular area including a part group area whose mutual distance is equal to or less than a certain threshold. Thereby, the part area | region when a pedestrian forms the group can be extracted appropriately.

  In the present embodiment, the score of the part group area is calculated as the sum of the scores of the high score area, and the score of the parts area is calculated as the sum of the scores of the part group area (S150 in FIG. 2). For example, in FIG. 6, the scores of the part group regions G1 and G2 are calculated as the sum of the scores of the high score areas TT included in the part group regions G1 and G2, and the sum of the scores of the part group regions G1 and G2 is calculated. Calculate the score. Thereby, the score of a parts area can be calculated comparatively easily.

  Furthermore, in this embodiment, when there is a part region P extracted based on the head part model as shown in FIG. The size of the body part is estimated from the size of the included high score area, and the pedestrian area S is estimated. That is, an object area where a target object exists (a pedestrian area referred to in the embodiment, and so on) is estimated based on the size of the high score area included in the part area. Thereby, a pedestrian area can be estimated from each part area.

  In addition, as shown in FIG. 7B, when there are two part areas P1 and P3 extracted based on the head part model, and there is a part area P2 extracted based on the leg part model, A pedestrian area is estimated based on the arrangement relationship. That is, the object area where the object exists is estimated based on the positional relationship between the part areas. Thereby, a more suitable pedestrian area can be estimated.

  At this time, as shown in FIG. 7C, when the two parts areas P1 and P2 are in a correspondence relationship, a circumscribed rectangle for the pedestrian areas S1 and S2 estimated from the parts areas P1 and P2 is set as a pedestrian. Estimated as region S3. That is, the object area is estimated as a circumscribed rectangle including the part area. Thereby, a more suitable pedestrian area can be estimated.

  And in estimation of a pedestrian area, when the sum total of the score of the parts area included in a pedestrian area is below a fixed threshold value, the said pedestrian area is excluded from extraction object. That is, the score of the object region based on the part region is calculated, and the object region whose score is equal to or less than a certain threshold is excluded from the extraction target. Thereby, a more reliable pedestrian area can be estimated.

  In this embodiment, when the pedestrian area is estimated, the collision risk determination unit 52 determines the collision risk based on the distance to the pedestrian output from the distance measuring device 31. Based on the degree of collision risk, notification to the driver via the speaker 41 by the notification control unit 70 of the vehicle control device 20 and collision avoidance control via the brake 42 and the steering 43 by the vehicle control unit 80 are performed. Is called. Thereby, driving assistance with higher safety is realized.

As described above, the present invention is not limited to the embodiment described above, and can be implemented in various forms without departing from the technical scope thereof.
(A) In the above embodiment, as shown in FIG. 5A, the high score areas TC and TT are grouped according to the distance and the high score areas TC and TT.

  (A-1) On the other hand, as shown in FIG. 5B, the part group region may be extracted only by the distance between the high score areas TC and TT. That is, the part group region may be extracted so as to include a high score area whose mutual distance is equal to or less than a certain threshold. In this way, the part group area can be extracted by a relatively simple process.

  (A-2) Further, as shown in FIG. 5C, the density is calculated by dividing the number of areas TT included in the part group regions G4 and G5 by the area of the part group regions G4 and G5. When the density exceeds a certain threshold, the part group region may be used, and when the density is less than the certain threshold, the part group region may be excluded. That is, when the density of the high score area in the part group area is equal to or less than a certain threshold value, the part group area may be excluded from the extraction target. In this way, the possibility of erroneous detection is reduced, and the part group region can be extracted more accurately.

  (A-3) Furthermore, in the example shown in FIG. 5C, the sum of the scores of the high score area TT included in the part group regions G4 and G5 is obtained, and the sum of the scores exceeds a certain threshold value. In some cases, a part group region may be used, and if it is below a certain threshold, it may be removed from the part group region. That is, when the sum of the scores of the high score area in the part group area is equal to or less than a certain threshold value, the part group area may be excluded from the extraction target. Even in this case, the possibility of erroneous detection is reduced, and the part group region can be extracted more accurately.

(B) In the above embodiment, the sum of the scores of the part group regions included in the part region is used as the score of the part region (S150 in FIG. 2).
(B-1) Here, in calculating the score of the part group region, the score may be calculated including the “positional proximity” of the high score area. Specifically, as shown in the lower part of FIG. 6, it is conceivable to define “closeness of position” of the high score area as the reciprocal of the difference sum of the positions p of the high score area. That is, when calculating the score of the part group region, the score may be calculated in consideration of the distance between the high score areas included in the part group region. In this way, the score of the parts area becomes more reliable.

  (B-2) Further, in calculating the score of the part group area, in addition to or instead of “closeness of position” of the high score area, it is assumed that the part group area includes high score areas of different sizes. The score may be calculated including “closeness of size”. Specifically, as shown in the lower part of FIG. 6, the “closeness of size” of each area can be defined as the reciprocal of the difference sum of the size s of the high score area. That is, when calculating the score of the part group region, the score may be calculated in consideration of the size difference of the high score area included in the part group region. In this way, the score of the parts area becomes more reliable.

  (B-3) Furthermore, in addition to or in place of “closeness of position” and “closeness of size”, the score of the part group area may be calculated including the density of the high score area in the part group area. . That is, when calculating the score of the part group region, the score may be calculated in consideration of the density of the high score area in the part group region. In this way, the score of the parts area becomes more reliable.

  (C) In the above embodiment, when the sum of the scores of the parts areas included in the pedestrian area (object area) is equal to or less than a predetermined threshold, the pedestrian area is excluded from the extraction target. On the other hand, when the score of the part area is calculated, if the score is equal to or less than a certain threshold value, the part area may be excluded from the extraction target. In this way, the pedestrian area is not estimated based on the part area having relatively small reliability, and the pedestrian area can be estimated more appropriately.

  (D) In the above embodiment, the collision risk is determined based on the distance to the pedestrian output from the distance measuring device 31. On the other hand, in addition to or instead of the distance to the pedestrian, the risk of collision may be determined using at least one of the position, size, and score of the pedestrian area (object area).

  (E) In the above embodiment, the distance to the pedestrian is measured by the distance measuring device 31, but the distance may be measured based on the image information from the imaging device 32 instead of the distance measuring device 31. . That is, the distance measuring device 31 may be omitted. Of course, by providing the distance measuring device 31, highly accurate measurement is possible.

  DESCRIPTION OF SYMBOLS 10 ... Image processing device, 20 ... Vehicle control device, 31 ... Distance measuring device, 32 ... Imaging device, 41 ... Speaker, 42 ... Brake, 43 ... Steering, 50 ... Control part, 51 ... Pedestrian detection part, 52 ... Collision Risk level determination unit, 60 ... storage unit, 61 ... parts model storage unit, 62 ... collision risk level determination table storage unit, 70 ... notification control unit, 80 ... vehicle control unit

Claims (23)

An image processing device (10) for detecting an object based on an image from an imaging device (32) mounted on a vehicle,
A part model storage means (61) for storing a plurality of part models capable of detecting parts as parts of the object from the image;
Using the part model stored in the part model storage means, a part area extracting means (51) for extracting a part area for each part;
Object area estimation means (51) for estimating an object area where the object exists based on the parts area extracted by the parts area extraction means;
An image processing apparatus comprising: The image processing apparatus according to claim 1.
The image processing apparatus, wherein the object is a pedestrian. The image processing apparatus according to claim 1 or 2,
The part region extraction means moves the part model in the image, calculates a degree of coincidence with an area that is an area overlapping with the part model as a score, and the score exceeds a predetermined set threshold. An image processing apparatus, wherein the part region is extracted based on a score area (S120). The image processing apparatus according to claim 3.
The image processing apparatus, wherein the setting threshold is determined for each part. The image processing apparatus according to claim 3 or 4,
The part region extracting means extracts a part group region including the high score area (S130), and then extracts the part region based on the part group region. The image processing apparatus according to claim 5.
The image processing apparatus, wherein the part region extraction unit extracts the part group region so as to include the high score area whose mutual distance is equal to or less than a predetermined threshold. The image processing apparatus according to claim 6.
The image processing apparatus, wherein the part area extraction unit extracts the part group area so as to include the high score areas having the same size. The image processing apparatus according to claim 6 or 7,
The image processing apparatus according to claim 1, wherein the part region extraction unit excludes the part group region from extraction targets when the density of the high score area in the part group region is equal to or less than a predetermined threshold. In the image processing device according to any one of claims 6 to 8,
The image processing apparatus according to claim 1, wherein the part region extraction unit excludes the part group region from the extraction target when the total score of the high score areas in the part group region is equal to or less than a predetermined threshold value. In the image processing device according to any one of claims 5 to 9,
The parts area extracting means extracts the parts area as a circumscribed rectangular area including the parts group area (S140).
An image processing apparatus. The image processing apparatus according to claim 10.
The image processing apparatus according to claim 1, wherein the part region extraction unit extracts the part region as a circumscribed rectangular region including the part group region whose mutual distance is equal to or less than a predetermined threshold. The image processing apparatus according to claim 10 or 11,
The part area extraction means calculates a score of the part group area included in the circumscribed rectangular area as a sum of scores of the high score area included in the part group area, and calculates a sum of scores of the part group areas. Calculating a part area score which is a score of the part area (S150);
An image processing apparatus. The image processing apparatus according to claim 12.
The said part area | region extraction means calculates a score in consideration of the distance of the said high score area contained in the said part group area | region, when calculating the score of the said part group area | region. The image processing apparatus according to claim 12 or 13,
The said part area | region extraction means calculates a score in consideration of the difference of the size of the said high score area contained in the said part group area | region, when calculating the score of the said part group area | region. In the image processing device according to any one of claims 12 to 14,
The image processing apparatus according to claim 1, wherein when calculating the score of the part group area, the part area extraction unit calculates the score in consideration of the density of the high score area in the part group area. In the image processing device according to any one of claims 12 to 15,
The image processing apparatus according to claim 1, wherein the part area extraction unit excludes the part area from an extraction target when the part area score of the part area is equal to or less than a predetermined threshold. In the image processing device according to any one of claims 3 to 16,
The object area estimation means estimates an object area where the object exists based on the size of the high score area included in the parts area (S160).
An image processing apparatus. In the image processing device according to any one of claims 3 to 17,
The object area estimation means estimates an object area where the object exists based on a positional relationship between the part areas (S160).
An image processing apparatus. The image processing apparatus according to claim 18.
The object area estimation means estimates the object area as a circumscribed rectangle including the part area (S160).
An image processing apparatus. In the image processing device according to any one of claims 1 to 19,
The object region estimation unit calculates a score of the object region based on the part region, and excludes an object region whose score is equal to or less than a predetermined threshold from an extraction target. In the image processing device according to any one of claims 1 to 20,
Collision risk determination means (52) for determining the collision risk based on at least one of the position, size, score, and distance to the object estimated by the object region estimation means An image processing apparatus, further comprising: An image processing device according to claim 21;
A vehicle control device (20) having notification control means (70) for notifying the driver based on the collision risk determined by the collision risk determination means;
A vehicle control system comprising: An image processing device according to claim 21;
A vehicle control device (20) having vehicle control means (80) for performing collision avoidance control based on the collision risk determined by the collision risk determination means;
A vehicle control system comprising: