JP2012173230A - Body recognizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a body recognizing device capable of recognizing a body of a recognition object from a pattern of a reflection point group received by an onboard radar, at a low cost without additionally installing another sensor such as an onboard camera.SOLUTION: A body recognizing device separates, by using a separation processing unit 6, an L-shaped pattern portion bent in a direction opposite to an own vehicle 1 out of a pattern of a reflection point group received by a laser radar 2, and recognizes a body of a recognition object by using a body attribute estimating unit 8 based on a rectangle of which one part is constituted by the L-shaped pattern portion separated by the separation processing unit 6.

Description

  The present invention relates to an object recognition apparatus for recognizing an object such as a vehicle, a bicycle, or a person (pedestrian) from a pattern of reflection points received by an in-vehicle radar.

  Conventionally, various preventive safety technologies have been researched and developed in the field of vehicles, and various preventive safety systems such as a front collision damage reduction system and a rear side warning system are being put into practical use.

  In these systems, if an object such as a vehicle, a bicycle, or a pedestrian can be distinguished and recognized from the pattern of reflection points received by an on-vehicle radar represented by a laser radar, the applicable range of the preventive safety system can be expanded at a low cost. .

  Then, it has been proposed to distinguish and recognize an object depending on which of the set three types of patterns L, I, and O the reflection point group pattern received by the laser radar as the vehicle-mounted radar belongs to ( For example, refer nonpatent literature 1).

Stephan Wender, 3 others, “Classification of Laserscanner Measurement at Interstitial Measurement at Interstitial Measurement in Interstitial Stimulation Measurement.” 2005. Proceedings. IEEE, IEEE, June 2005, p. 94-99

  As described in Non-Patent Document 1, when recognizing and recognizing an object from the pattern of the reflection point group received by laser radar, the object, vehicle, motorbike, bicycle, pedestrian, or the like that is the object to be recognized is regarded as a rectangle, and the reflection is performed. Assuming that the point cloud pattern forms a part of a rectangle, it is conceivable that an object is recognized by estimating (attribute estimation) whether the rectangle belongs to a rectangle of a vehicle, a motorbike, a bicycle, or a pedestrian.

  FIG. 7 shows an example of object recognition based on the attribute estimation. In this case, the laser radar 101 of the host vehicle 100 scans the front, and the laser pulse output from the laser radar 101 is a part of each of the vehicle 200 and the motorbike 300 to be recognized in front (left side and rear end in FIG. 7). The reflection points r of the vehicle 200 and the motorbike 300 along these edges are received by the laser radar 101 and detected. Then, the reflection points r that are closer together by the clustering process are combined into one cluster to form a reflection point group.

  At this time, if the vehicle 200 and the motorbike 300 are separated from the host vehicle 100, the reflection point group of the vehicle 200 and the reflection point group of the motorbike 300 become separate cluster Ca and cluster Cb reflection point groups. The vehicle 200 and the motorbike 300 in front of the host vehicle 100 can be recognized from the attribute estimation of the rectangles Ra and Rb based on the L-shaped patterns of the reflection points of Ca and Cb.

  However, if the vehicle 200 and the motorbike 300 are close to each other when viewed from the host vehicle 100, attribute estimation becomes difficult and the vehicle 200 and the motorbike 300 cannot be recognized, as will be described below.

  FIG. 8 shows an example in which attribute estimation becomes difficult. In this case, the vehicle 200 and the motorbike 300 approach each other when viewed from the host vehicle 100, and the reflection points r of the vehicle 200 and the reflection points r of the motorbike 300 are continuously connected. r is collected into a pattern of reflection point groups of one cluster Cc. At this time, since the rectangle Rc based on the pattern of the reflection point group of the cluster Cc does not correspond to any rectangle of a vehicle, a motorbike, a bicycle, or a pedestrian, it is difficult to estimate the attribute of the recognition target object from the rectangle Rc. The vehicle 200 and the motorbike 300 cannot be recognized.

  As described above, the situation in which the attribute estimation of the recognition target object becomes difficult and the recognition cannot be performed is not only when the plurality of recognition target objects are close to each other, for example, the recognition target object and other objects ( This also occurs when, for example, a guard rail or the like approaches and the reflection points are continuously connected.

  In addition, the host vehicle 100 further includes another sensor such as an in-vehicle camera, and each sensor to be recognized is detected by the sensor, and each object to be recognized is clustered when the reflection point group received by the laser radar 101 is clustered. As a group of reflection points in separate clusters, recognition is possible even when multiple objects to be recognized are close to each other or when the object to be recognized is close to other objects. However, in this case, it is necessary to further include another sensor such as a vehicle-mounted camera, which is expensive and impractical.

  An object of the present invention is to enable recognition of an object to be recognized from a pattern of a reflection point group received by an in-vehicle radar at a low cost without further providing another sensor such as an in-vehicle camera. In the present invention, the “L-shaped portion” includes an L-shaped portion as viewed from the host vehicle and an inverted L-shaped portion obtained by horizontally inverting the L-shape.

  In order to achieve the above object, an object recognition apparatus of the present invention is an object recognition apparatus that recognizes an object based on a pattern of a reflection point group received by an in-vehicle radar, and is opposed to the vehicle from the pattern. Separating means for separating an L-shaped portion bent in a direction, and recognition means for recognizing an object from a rectangle that constitutes a part of the L-shaped portion separated by the separating means. (Claim 1).

  According to the invention of claim 1, focusing on the fact that the part of the object to be recognized in the pattern of the reflection point group received by the in-vehicle radar is L-shaped bent in the opposite direction to the host vehicle, the separating means The corresponding L-shaped part is separated from the pattern of the reflection point group received by the in-vehicle radar. Furthermore, an object such as a vehicle, a motorbike, a bicycle, a pedestrian, etc. is regarded as a rectangle, and each object to be recognized can be recognized by the recognition means based on a rectangle in which each separated L-shaped part constitutes a part. it can.

  In this case, even if the reflection point group pattern includes reflection points of a plurality of objects to be recognized, the reflection points of each object form different L-shaped parts, so that these L-shaped parts are separated. By doing so, each object to be recognized can be reliably recognized as one object (same object) based on the pattern of each L-shaped portion without being erroneously recognized.

  In addition, even if the reflection point group pattern includes reflection points of a recognition target object and other objects that are not recognition targets, the reflection point of the recognition target object is changed to an L-shaped portion from the reflection point group pattern. And the object to be recognized can be reliably recognized based on the pattern of the separated L-shaped part.

  Therefore, it is possible to reliably separate the object part to be recognized from the reflection point cloud pattern received by the in-vehicle radar at a low cost without further providing other sensors such as an in-vehicle camera. Can do.

It is a block diagram of one embodiment of an object recognition device of the present invention. It is explanatory drawing of the recognition process of FIG. It is a flowchart for operation | movement description of FIG. It is explanatory drawing of an example of the pattern of the reflective point group sent to the isolation | separation processing part from the clustering process part of FIG. (A) is explanatory drawing of the other example of the pattern of the reflective point group sent to the isolation | separation processing part from the clustering process part of FIG. 1, (b) is the one part enlarged view. It is explanatory drawing of isolation | separation and recognition of the further another example of the pattern of the reflective point group sent to the isolation | separation processing part from the clustering process part of FIG. It is explanatory drawing of the recognition process in case the pattern of the reflection point group of each of the some object of recognition object is obtained. It is explanatory drawing of the conventional recognition process in case one pattern of the reflective point group containing the several object of recognition object is obtained.

  An embodiment of the object recognition apparatus of the present invention will be described with reference to FIGS.

  FIG. 1 shows a configuration of an object recognition device provided in the own vehicle 1, and this object recognition device includes a laser radar 2 as an in-vehicle radar. The laser radar 2 is a wide-angle ranging radar that searches a predetermined range in front of the host vehicle 1 (for example, a range of about 180 °). By using near-infrared rays, reflection from a bicycle or a pedestrian can be obtained. It is done. The laser radar 2 is attached to, for example, a front bumper of the host vehicle 1 and outputs a laser pulse while exploring the front of the host vehicle 1 for each frame at intervals of, for example, 80 ms. Reflection on an object such as a motorcycle, bicycle, pedestrian (person), etc. is received, and position data (distance from the vehicle 1 and position data in the left-right direction) of each reflection point is output to the reception processing unit 3. The laser radar 2 actually scans the front, up, down, left, and right, but here, the laser radar 2 searches in the left-right direction for the sake of simplicity.

  The reception processing unit 3 receives the position data of each reflection point of the laser radar 2 and sends it to the clustering processing unit 5 of the recognition processing unit 4 having a microcomputer configuration.

  FIG. 2 schematically shows an object recognition processing process of the object recognition apparatus, and outputs a pulse while scanning the front of the own vehicle 1 from the laser radar 2 of the own vehicle 1 as in the case described with reference to FIGS. The reflected laser is reflected at the edges of a part of the vehicle 200 and the motorbike 300 (the left side and the rear in FIG. 2) which are objects to be recognized in front, and the reflection at these edges is received by the laser radar 2 Each reflection point r along the edge of each of the vehicle 200 and the motorbike 300 is detected, and the reception processing unit 3 sends the position data of each reflection point r to the clustering processing unit 5 in FIG.

  Based on the position data of each reflection point r, the clustering processing unit 5 connects the reflection points r in the vicinity using, for example, a well-known morphological operation, and sets the reflection points r in the vicinity as a cluster (cluster of reflection points). Perform clustering processing collectively. The reflection point group pattern P1 of the cluster C1 of FIG. 2 formed by this clustering processing is the reflection point group pattern received by the in-vehicle radar of the present invention.

  The pattern P1 of the reflection point group in FIG. 2 is that the vehicle 200 to be recognized and the motorbike 300 that are obstacles approach each other as viewed from the own vehicle 1, so that the reflection point r is the reflection point group of the cluster C1. Concatenated as included. When the vehicle 200 and the motorbike 300 are separated from the host vehicle 1, as described in FIG. 7, the reflection point cloud pattern of the vehicle 200 and the reflection point cloud of the motorbike 300 are processed by the clustering process. The pattern is formed separately.

  Next, the reflection point group pattern P1 of the clustering processing unit 5 is sent to the separation processing unit 6 of FIG. 1 forming the separation means of the present invention. The separation processing unit 6 separates L-shaped portions L1 and L2 bent in the opposite direction with respect to the host vehicle 1 from the pattern P1.

  That is, when the pattern P1 includes a plurality of objects to be recognized such as the vehicle 200 and the motorbike 300, the host vehicle 1 normally detects each object in the right hand front (or left hand front). The reflection point of the laser at is the reflection point of the edge of the two sides of the L shape (or the inverted L shape reversed left and right). In this case, the reflection point group of each object forms an L-shaped portion bent in the opposite direction with respect to the own vehicle 1 such as the L-shaped portions L1 and L2 of the pattern P1. Therefore, the separation processing unit 6 separates the L-shaped portions L1 and L2 of the pattern P1 as the reflection point group patterns of the object (patterns P11 and P12 in FIG. 2). This separation method will be described later. If the pattern P1 is a pattern of one object to be recognized, there will be only one L-shaped part, so there will be only one pattern after separation.

  The patterns P11 and P12 separated by the separation processing unit 6 are written and held in the storage unit 7 of FIG. 1 having an identification label for tracking over a plurality of set frames and configured by a RAM, a flash memory, or the like. At the same time, it is sent to the object attribute estimation unit 8 of FIG. 1 forming the recognition means of the present invention.

  The object attribute estimation unit 8 tracks the patterns P11 and P12 over a plurality of frames in consideration of the relative movement of the host vehicle 1 and each of the objects ahead in order to eliminate erroneous separation results due to disturbance or the like. Then, on the condition that the same patterns P11 and P12 are detected over the plurality of frames, it is determined that the patterns P11 and P12 are the patterns of the objects to be recognized, and each of the patterns P11 and P12 is L-shaped. The rectangles R11 and R12 of FIG. 2 having two sides (parts) of the shape (or inverted L shape) are drawn and formed. Further, for example, by the pattern matching between the formed rectangles R11 and R12 and the rectangles set for each recognition target object such as a vehicle, a motorbike, a bicycle, and a pedestrian, which rectangle the rectangles R11 and R12 belong to And the attributes of the objects of the rectangles R11 and R12 (whether they belong to a vehicle, a motorbike, a bicycle, a pedestrian, etc.) are estimated and recognized. And an estimation result is sent to the driving assistance part 9 of FIG.

  The driving support unit 9 grasps from the recognition result of the object attribute estimation unit 8 that the vehicle 200 and the motorbike 300 exist as obstacles in front of the host vehicle 1, and provides driving support for preventive safety according to the recognition result. Do.

  Steps S1 to S6 in FIG. 3 show an outline processing procedure of the object recognition apparatus. First, the reception processing unit 3 acquires position data of each reflection point r of the laser radar 2 (step S1), and based on the data. Then, each reflection point r is clustered by the clustering processing unit 5 (step S2).

  Next, the separation processing unit 6 separates the L-shaped portions L1 and L2 that are bent in the opposite direction with respect to the host vehicle 1 from the reflection point group pattern P1 of the processing result of the clustering processing unit 5.

  Then, the object attribute estimation unit 8 tracks the patterns P11 and P12 of the L-shaped portions L1 and L2 over a plurality of frames (step S4), and the patterns P11 and P12 of the L-shaped portions L1 and L2 are traced by this tracking. If the pattern of the object to be recognized is confirmed, the attributes of the objects of the rectangles R11 and R12 of the patterns P11 and P2 of the L-shaped portions L1 and L2 are estimated to recognize the object (step S5).

  Further, based on the recognition result of the object attribute estimation unit 8, the driving support unit 9 performs necessary driving support.

  Next, a separation method of the separation processing unit 6 will be described.

  When separating the L-shaped portions L1 and L2 of the pattern P1 into the reflection point group patterns (patterns P11 and P12 in FIG. 2) of the object (vehicle 200, motorbike 300), respectively, for example, an L-shaped reference Although it is conceivable to perform pattern matching using a pattern, in this embodiment, the connecting portions of the L-shaped portions L1 and L2 of the pattern P1 (portions surrounded by broken lines in FIG. 2) are recessed when viewed from the own vehicle 1. When the reflection point r is traced from either end of the pattern P1, the convex portion of the bending point of the L-shaped portion L1 or L-shaped portion L2 (enclosed by a one-dot broken line in FIG. 2) Part) Focusing on reaching the concave portion α through β, the pattern P is cut at the concave portion (connecting portion) α and separated into patterns P11 and P12 as described below.

  That is, on the assumption that the object to be recognized is a rectangle, the pattern P1 of the cluster C1 of the reflection point group of the laser radar 2 is concave if it is a pattern of one object (for example, the vehicle 200 or the motorbike 300). Although the portion α does not exist, if the pattern includes a plurality of objects (for example, the vehicle 200 and the motorbike 300) that are close to each other, a concave portion α exists at the boundary between the patterns (for example, the patterns P11 and P12) of each object. There is a convex portion β of an L-shaped (or reverse L-shaped) bent.

  Therefore, the separation processing unit 6 traces the reflection point r from one end or the other end of the pattern P1, changes the direction of the trajectory by a predetermined angle or more, and converts the portion where the trajectory is bent to the opposite side of the own vehicle 1 to the convex portion β. A candidate position β * is detected, and then a part that is bent by tracing the reflection point r and changing the trajectory by a predetermined angle or more is detected as a candidate position α * of the concave portion α.

  FIG. 4 shows the candidate positions α * and β * in the pattern P1 of the vehicle 200 and the motorbike 300 that are close to each other, and one end side (for example, the upper end portion in FIG. 4) of the pattern P1 and the candidate position α * of the concave portion α. There is a candidate position β * of the convex portion β of the motorbike 300 between the other end side (for example, the lower end portion of FIG. 4) of the pattern P1 and the candidate position α * of the concave portion α. There is a candidate position β * of the convex portion β. In this case, the trajectory is traced from the detected candidate position β * of the convex portion β of the vehicle 200 or the candidate position β * of the convex portion β of the motorbike 300, and the portion that is bent by changing by a predetermined angle or more is the concave portion α. It is detected as a candidate position α *.

  Furthermore, as shown in FIG. 4, a straight line distance (length) from the detected candidate position β * to the candidate position α * of the concave portion α that reaches the reflection point r by tracing the reflection point r is set as Wx, and the determination threshold value is set. , Wx ≧ W0, the detected candidate position α * is determined as the position of the recess α, and the pattern P1 is cut at the position of the recess α to obtain patterns P11 and P12. To separate.

  Then, by separating the pattern P1 into the pattern P11 of the vehicle 200 and the pattern P12 of the motorbike 300, one pattern P1 includes the reflection points r of a plurality of objects to be recognized (the vehicle 200 and the motorbike 300). Even so, the pattern P1 can be separated into the patterns P11 and P12 of each object, and each object can be individually recognized based on the attribute estimation of the respective rectangles R11 and R12.

  Even when the reflection point r of three or more objects to be recognized is included in one pattern P1, the concave portion α at the boundary of each object is detected in the same manner, and the pattern P1 is detected as each concave portion α. By separating in (5), it is possible to obtain a pattern of each separated object, obtain each rectangle, and individually recognize each object to be recognized based on attribute estimation of each rectangular object.

  The condition that Wx ≧ W0 is that, for example, if the pattern P1 includes a bulging portion such as a door mirror of the vehicle 200, the same concave and convex portions are generated in the pattern P1. This is to prevent separation in the part of.

  FIG. 5A shows a case where the pattern of the reflection point group of the clustering processing unit 5 is the reflection point group pattern P2 of the front end and the right edge of the single vehicle 200. At this time, the pattern P2 is the door mirror of the vehicle 200. Assuming that the protrusion 200a is small but protrudes in the portion 200a, the locus following the reflection point r of the pattern P2 is a protruding end of the door mirror 200a as shown in FIG. There is a possibility that the bent portion is changed by a predetermined angle or more at (a portion surrounded by a broken line γ in the figure), and this portion is erroneously detected as a convex portion β. Further, based on this erroneous detection, a portion where the locus is bent by changing by a predetermined angle or more following the reflection point r (a portion between two positions surrounded by a broken line δ in the figure) is a concave portion α. As a result, the pattern P2 is erroneously separated at the broken line δ.

  Here, the linear distance ωx between the broken line γ portion and the broken line δ portion in FIG. 5B is the concave portion α reaching the reflection point r from the candidate position β * of the convex portion β in FIG. It is sufficiently smaller than the linear distance Wx to the candidate position α *.

  Therefore, when the above-described determination threshold value W0 is set to W0> ωx, and there is a protruding portion such as the door mirror 200a on the object to be recognized, that portion of the broken line δ in FIG. Avoid misdetection as α.

  Next, even if the pattern of the reflection point group of the laser radar 2 is the pattern of the reflection point of one object to be recognized, the pattern includes a reflection point outside the recognition target such as a roadside guardrail, for example. For this, it is necessary to separate the patterns.

  FIG. 6 shows a case where the recognition target vehicle 200 is stopped by approaching the guard rail 400 in the direction in which the host vehicle 1 travels. In this case, the reflection point r between the vehicle 200 and the guard rail 400 is determined by clustering processing of the reflection point r. A reflection point group pattern P3 is obtained. The pattern P3 includes a bent convex portion β of the L-shaped portion L3 based on the reflection point r of the front end and the left edge of the vehicle 200, and a concave portion α at the boundary between the vehicle 200 and the guard rail 400.

  Then, the separation processing unit 6 detects the candidate position of the convex portion β by tracing the reflection point r of the pattern P3 from the end on the vehicle 200 side by the above-described separation processing, and further, the reflection point r of the pattern P3. Then, the candidate position of the concave portion α is detected, this candidate position is determined as the concave portion α, the pattern P3 is cut by the concave portion α, and the pattern of the vehicle 200 similar to the pattern P11 as shown in FIG. It separates into P31 and the pattern P32 of the guardrail 400. The linear distance Wy between the convex portion β and the concave portion α in FIG. 6 is also Wy> W0. Further, when the reflection point r of the pattern P3 is traced from the end on the guard rail 400 side, even if the convex portion β is detected, the subsequent concave portion α cannot be detected. In such a case, the reflective point of the pattern P3 After r is traced to the end on the vehicle 200 side, the reflection point r of the pattern P3 is traced again from the end on the vehicle 200 side to detect the convex portion β and the subsequent concave portion α to determine the concave portion α.

  Accordingly, even when the pattern P3 includes reflection points of an object to be recognized (for example, the vehicle 200) and another object (for example, a guardrail), the pattern P31 of the object to be recognized is separated and the object is rectangular R31. Can be recognized based on the attribute estimation.

  As described above, according to the present embodiment, a case where a plurality of objects are included in the reflection point cloud patterns P1, P2, and P3 received by the laser radar 2, or an object to be recognized and other objects are detected. Even if included, the L-shaped portions L1, L2, L3 of each object to be recognized are separated into individual patterns P11, P12, P31 from the patterns P1, P2, P3, Without further detecting the sensor, it is possible to reliably recognize each object to be recognized by performing attribute estimation at low cost.

  The present invention is not limited to the above-described embodiment, and various modifications other than those described above can be made without departing from the spirit of the present invention. Of course, it is not limited to a radar, and may be a millimeter wave radar, an ultrasonic radar, or the like. The search range of the in-vehicle radar may be any, and the in-vehicle radar may search for the rear and left and right sides of the host vehicle 1 and search for the entire circumference of the host vehicle 1. May be. For example, the present invention can be similarly applied to a case where the rear of the host vehicle is searched by an in-vehicle radar to recognize an object behind the host vehicle.

  Next, the recognition target object may be a part of the above-described vehicle, motorbike, bicycle, pedestrian, or the like, or may be an object other than these.

  Next, the attribute estimation method of the object attribute estimation unit 8 based on the pattern separated by the separation processing unit 6 may be any method, and is not limited to the estimation method of the embodiment.

  And this invention is applicable to the object recognition apparatus of the various vehicles provided with the vehicle-mounted radar.

DESCRIPTION OF SYMBOLS 1 Own vehicle 2 Laser radar 3 Reception process part 4 Recognition process part 5 Clustering process part 6 Separation process part 8 Object attribute estimation part L1, L2, L3 L-shaped part P1, P2, P3, P11, P12, P31, P32 pattern R11, R12, R31 Rectangular r Reflection point

Claims (1)

An object recognition device for recognizing an object based on a pattern of reflection points received by an on-vehicle radar,
Separating means for separating an L-shaped portion bent in the opposite direction with respect to the host vehicle from the pattern;
An object recognition apparatus comprising: a recognition means for recognizing an object from a rectangle in which the L-shaped part separated by the separation means constitutes a part.

Images