JP2012145444A - Pedestrian detection device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a pedestrian detection device which detects a pedestrian from detection point groups with a high degree of accuracy.SOLUTION: A pedestrian detection device which groups detection points detected by detection means into a plurality of detection point groups D4 and detects a pedestrian from the same comprises: moving direction calculation means S5 to detect a moving direction (corresponding to the direction of a velocity vector D5) of each of the detection point groups D4; distribution information calculation means S7 to calculate distribution information D7 on distribution of each of the detection point groups D4 with respect to the moving direction; and determination means S9 to determine whether or not a temporal change in the distribution information D7 on each of the detection point groups D4 with respect to the moving direction has periodicity and to specify the detection point group having the periodicity as the pedestrian.

Description

  The present invention relates to a pedestrian detection apparatus that groups detection points detected by detection means as a plurality of detection point groups and detects a pedestrian from these detection point groups.

  The driver of the vehicle must pay attention to pedestrians and the like existing around the vehicle while traveling. In order to reduce such a burden on the driver, an apparatus for detecting a pedestrian around the vehicle has been developed. As a pedestrian detection device, for example, in Patent Document 1, using a time-series data of three-dimensional position information of detection points around a vehicle acquired by a laser radar, a distribution of point groups on a plane along the time series The periodicity of the change is analyzed, and a point cloud with periodicity is detected as a pedestrian. Moreover, in patent document 1, the detection point of the height equivalent to the leg part vicinity of a pedestrian is acquired as a point cloud, and the size of the point cloud on a plane falls within a fixed range is set as a pedestrian candidate point cloud. To improve accuracy.

JP 2010-205042 A JP 2009-288099 A JP 2004-185363 A

  In the case of the pedestrian detection device of Patent Document 1, an object having periodicity may be erroneously detected as a pedestrian due to other than the movement of the pedestrian. For example, if a periodic distribution change occurs in a point cloud due to a laser radar measurement error or the like, the point cloud may be erroneously detected as a pedestrian. Further, in the case of another moving body (for example, a bicycle) in which the distribution change of the point cloud has the same periodicity as that of a pedestrian, there is a possibility that it is erroneously detected as a pedestrian. In addition, in the case of a stationary object having a periodic movement (for example, a banner or a tree that swings in the wind), it may be erroneously detected as a pedestrian. Furthermore, in the case of the pedestrian detection device of Patent Document 1, since the stride increases when the pedestrian runs, the size of the point cloud increases, and when the size exceeds a certain range, the point cloud is detected from the pedestrian candidate. It is excluded and cannot be detected as a pedestrian.

  Then, this invention makes it a subject to provide the pedestrian detection apparatus which detects a pedestrian from a detection point group with high precision.

  A pedestrian detection device according to the present invention is a pedestrian detection device that groups detection points detected by a detection unit as a plurality of detection point groups and detects a pedestrian from these detection point groups, the detection point group A moving direction calculating unit that calculates a moving direction of the detection point group for each detection point; a distribution information calculating unit that calculates distribution information of the detection point group with respect to the moving direction calculated by the moving direction calculating unit for each detection point group; Determining means for determining whether the temporal change of the distribution information of the detection point group with respect to the moving direction calculated by the distribution information calculating means for each group has periodicity, and determining the detection point group having periodicity as a pedestrian It is characterized by providing.

  In this pedestrian detection device, each detection point for an object is acquired by a detection means, the acquired detection points are grouped and divided into a plurality of detection point groups, and a pedestrian is detected from the plurality of detection point groups. In particular, in the pedestrian detection device, for each detection point group, the movement direction calculation unit calculates the movement direction of the detection point group, and the distribution information calculation unit calculates distribution information (for example, the detection point group) of the detection point group. The size in the moving direction and the dispersion value in the moving direction of the detection point group) are calculated. In the case of a pedestrian, since it moves to some extent in a certain direction, it is because it also has periodicity due to the movement of the legs along the moving direction. Therefore, in the pedestrian detection device, for each detection point group, the determination unit determines whether or not the temporal change of the distribution information with respect to the movement direction of the detection point group has periodicity, and the temporal change of the distribution information with respect to the movement direction. A detection point group having periodicity is determined as a pedestrian. Thus, in the pedestrian detection apparatus, it is possible to detect a pedestrian with high accuracy by determining whether or not the temporal change of the distribution information with respect to the moving direction of the detection point group has periodicity. Thereby, it is possible to reduce erroneous detection of a detection point group for an object having periodicity due to other than the movement of a pedestrian.

  In the pedestrian detection device of the present invention, it is preferable that the determination unit determines that the detection point group is a pedestrian when there is a correlation specific to the pedestrian in the distribution information with respect to the movement speed and the movement direction of the detection point group. .

  In the case of a pedestrian, the stride is small when walking (when the speed is slow), and the stride is large when running (when the speed is fast). Thus, there is a correlation between the pedestrian speed and the stride. Therefore, in this pedestrian detection device, the determination means detects when there is a pedestrian-specific correlation between the movement speed of the detection point group and the distribution information for the movement direction (the distribution information for the movement direction depends on the stride). The point cloud is determined as a pedestrian. Thus, in the pedestrian detection device, it is possible to detect a pedestrian with higher accuracy by determining whether or not there is a correlation specific to the pedestrian in the distribution information with respect to the moving speed and the moving direction of the detection point group. Can do. For example, even if a periodic change occurs in the distribution of detection point groups due to detection errors of the detection means, etc., there is a possibility that there is a pedestrian-specific correlation in the distribution information for the movement speed and movement direction for the detection point group. Is extremely low, and it is possible to reduce erroneous detection of the detection point group. In addition, in the case of other moving bodies (for example, bicycles) whose temporal changes in the distribution information of the detection point group have the same periodicity as pedestrians, the distribution information for the moving direction does not change even when the speed is low or high. It is possible to reduce erroneous detection of such a moving body. Moreover, even when a pedestrian is walking or running, it can be detected as a pedestrian.

  In the pedestrian detection device of the present invention, the determination unit accumulates movement information of the detection point group for each detection point group, and the detection point group is stable for a certain period of time based on the accumulated movement information of the detection point group. If it is determined that the detected point group is moving, it is preferable to determine the detected point group as a pedestrian.

  The pedestrian moves continuously in a predetermined direction at a predetermined speed to some extent, and changes with time such as a moving direction and a moving speed are small. Therefore, in this pedestrian detection device, movement information (for example, movement direction, movement speed, and position) of the detection point group is accumulated in time series for each detection point group. In the pedestrian detection device, the determination unit moves the detection point group stably for a predetermined time based on the movement information of the detection point group for a predetermined time (that is, movement with little change in the movement information). When it is determined, the detected point group is determined as a pedestrian. Thus, in the pedestrian detection device, it is possible to detect the pedestrian with higher accuracy by determining whether or not the detection point group moves stably for a certain period of time. For example, in the case of a stationary object with periodic motion (for example, a flag or tree that swings in the wind), its moving direction and speed change randomly in a short time, so such a stationary moving object is erroneously detected. Can be reduced.

  According to the present invention, it is possible to detect a pedestrian with high accuracy by determining whether or not the temporal change of the distribution information with respect to the moving direction of the detection point group has periodicity.

It is a block diagram of the pedestrian detection apparatus which concerns on this Embodiment. It is an example of the time change of the clustering point group which concerns on this Embodiment, (a) is a clustering point group with respect to the vehicle which goes straight from the front direction, (b) is a clustering point group with respect to the vehicle which turns left from the front direction. (C) is a clustering point group for a pedestrian who goes straight from the front direction. It is a flowchart which shows the flow of the main process of the pedestrian detection which concerns on this Embodiment. It is a flowchart which shows the flow of the point cloud movement information determination process which concerns on this Embodiment. It is a flowchart which shows the flow of the variation amount normalization process which concerns on this Embodiment. It is a flowchart which shows the flow of the periodicity determination process which concerns on this Embodiment.

  Hereinafter, an embodiment of a pedestrian detection device according to the present invention will be described with reference to the drawings. In addition, the same code | symbol is attached | subjected about the element which is the same or it corresponds in each figure, and the overlapping description is abbreviate | omitted.

  In the present embodiment, the pedestrian detection device according to the present invention is applied to a pedestrian detection device mounted on a vehicle. The pedestrian detection apparatus according to the present embodiment measures a vehicle periphery by using a laser radar, acquires three-dimensional information of the measured detection points, and uses a plurality of moving body attributes (three-dimensional information using the detection points). For example, pedestrians are identified from among pedestrians, vehicles (passenger cars, large vehicles, etc.), bicycles, motorcycles). Information on the pedestrian detected by the pedestrian detection device is used in a driving support device that requires pedestrian information.

  With reference to FIG.1 and FIG.2, the pedestrian detection apparatus 1 which concerns on this Embodiment is demonstrated. FIG. 1 is a configuration diagram of a pedestrian detection device according to the present embodiment. FIG. 2 is an example of a time change of the clustering point group according to the present embodiment, where (a) is a clustering point group for a vehicle going straight from the front direction, and (b) is a clustering for a vehicle turning left from the front direction. It is a point cloud, and (c) is a clustering point cloud for a pedestrian who goes straight from the front direction.

  The pedestrian detection apparatus 1 separates a three-dimensional object from a three-dimensional point group of detection points measured by a laser radar, clusters the point group of the three-dimensional object, and identifies whether the person is a pedestrian for each clustered point group. To do. In this pedestrian identification, for each clustering point cloud, the point cloud is tracked using the time series data to calculate the velocity vector, and the amount of variation with respect to the movement direction of the point cloud (the direction of the velocity vector) is calculated. Then, based on the time series data of the variation amount, the presence / absence of periodicity of the variation amount in time is determined to identify the pedestrian. In particular, it is determined whether or not the point cloud has moved stably for a certain period of time based on the accumulated data of the point cloud movement information, and the amount of variation is normalized according to the moving speed (speed of the speed vector) The periodicity is determined using the variation amount. Note that each clustering point group (that is, each detection point group) refers to all detection point groups or some detection points (for example, by removing detection point groups that are too large). Even if it is a group, either may be sufficient.

  For this purpose, the pedestrian detection device 1 includes a laser radar 2 and an ECU [Electronic Control Unit] 3. In the present embodiment, the laser radar 2 corresponds to detection means described in the claims.

  Here, with reference to FIG. 2, the time change of the variation amount of the clustering point group will be described. In FIG. 2, detection points projected on a two-dimensional plane are indicated by black circles, and a moving direction of the clustering point group is indicated by a thick arrow. Further, FIG. 2 shows clustering point groups and moving directions at successive times t1, t2, t3, and t4.

  FIG. 2A shows a time change of a clustering point group composed of detection points for a vehicle traveling straight from the front direction of the host vehicle. In this case, the amount of variation at each time t1, t2, t3, t4 of the point group with respect to the movement direction D11, D12, D13, D14 at each time t1, t2, t3, t4 of the vehicle (in this example, the movement direction The lengths L11, L12, L13, and L14) of the point cloud hardly change even if the time changes.

  FIG. 2B shows a time change of a clustering point group including detection points for a vehicle turning left in the front direction of the host vehicle. In this case, the amount of variation at each time t1, t2, t3, t4 of the point cloud with respect to the moving direction D21, D22, D23, D24 at each time t1, t2, t3, t4 of the vehicle (of the point group in the moving direction). The lengths L21, L22, L23, and L24) monotonously increase with time.

  FIG. 2C shows a time change of the clustering point group composed of detection points for a pedestrian who goes straight from the front direction of the host vehicle. In this case, the amount of variation at each time t1, t2, t3, t4 of the point cloud with respect to the movement direction D31, D32, D33, D34 of the pedestrian at each time t1, t2, t3, t4 (point cloud in the movement direction). The lengths L31, L32, L33, and L34) periodically change due to the movement of the legs and hands (particularly legs) of the pedestrian. In other words, when one of the left and right legs is forward and the other is rearward, or when one is rearward and the other is forward, the lengths L32 and L34 of the moving direction point group become longer and forward. The lengths L31 and L33 of the point cloud in the moving direction are shortened when the leg that has moved back and the leg that has been behind moved forward. Further, in the case of a pedestrian, since the stride is small when walking, the length of the point cloud in the moving direction is shortened, and when the user is running, the stride is large, so the length of the point cloud in the moving direction is long.

  The laser radar 2 is a scanning radar that detects an object using laser light. The laser radar 2 is attached to a predetermined position of the vehicle in order to detect the periphery of the vehicle. The detection direction around the vehicle is at least the front of the vehicle and detects other directions as necessary. In the laser radar 2, in order to detect the horizontal direction at regular intervals, the light emitting unit and the light receiving unit are rotated in the horizontal direction, the laser beam is emitted at each horizontal scanning angle, and the reflected laser beam is received. To do. Further, in the laser radar 2, in order to detect the vertical direction at regular intervals, the light emitting unit and the light receiving unit are rotated in the vertical direction, the laser beam is emitted at each scanning angle in the vertical direction, and is reflected. Is received. In the laser radar 2, a radar including information (horizontal scanning angle, vertical scanning angle, emission time, light reception time, light reception intensity (reflection intensity), etc.) about each reflection point (detection point) that can be received. A signal is transmitted to ECU3. A configuration in which a plurality of laser radars are provided in the vertical direction or the horizontal direction and each laser radar scans only one direction in the horizontal direction or the vertical direction may be employed.

  The ECU 3 is an electronic control unit including a CPU [Central Processing Unit], a ROM [Read Only Memory], a RAM [Random Access Memory], and the like, and comprehensively controls the pedestrian detection device 1. The ECU 3 loads the application for the pedestrian detection device 1 stored in the ROM into the RAM and executes it by the CPU, thereby performing distance measurement processing, road plane estimation processing, three-dimensional object separation processing, clustering processing, and point cloud tracking. Processing, point cloud movement information determination processing, variation amount calculation processing, variation amount normalization processing, and periodicity determination processing are performed. In the present embodiment, the point cloud tracking process corresponds to the moving direction calculating means described in the claims, and the variation amount calculating process corresponds to the distribution information calculating means described in the claims. Point cloud movement information determination processing, variation amount normalization processing, and periodicity determination processing correspond to the determination means described in the claims.

  In the distance measurement processing, for each detection point that can be detected by the laser radar 2, a relative distance to the detection point is calculated based on the time from the emission of the radar light to the reception of the detection point, and further, the detection point is detected. Based on the horizontal scanning angle and the vertical scanning angle and relative distance, a three-dimensional position (front-rear direction position, lateral direction position) relative to the reference point of the vehicle (for example, the position where the laser radar 2 is attached, the center of gravity position of the vehicle). Position, height position). In the distance measurement process, a three-dimensional point group including three-dimensional position information of all detection points detected by the laser radar 2 is acquired. Furthermore, other information such as the relative speed of each detection point is also calculated as necessary.

  In the road plane estimation process, the detection point for the road plane is estimated from the detection points of the three-dimensional point group. As this estimation method, for example, it is determined whether or not the road plane is based on the height position of the detection point. Then, in the road plane estimation process, a road plane point group made up of three-dimensional position information of detection points on the road plane is acquired.

  In the three-dimensional object separation process, the detection points of the three-dimensional object are separated by excluding the detection points of the road plane point group from the detection points of the three-dimensional point group. In the three-dimensional object separation process, a three-dimensional object point group including three-dimensional position information of detection points of all three-dimensional objects is acquired.

  In the clustering process, detection points that can be regarded as one object are combined by comparing the three-dimensional positions of the detection points of the three-dimensional object point group, and the detection points included in the three-dimensional object point group are clustered. In the clustering process, a clustering point group including three-dimensional position information of a plurality of detection points divided by clustering is acquired.

  In the point cloud tracking process, the clustering point cloud for the same object is tracked for each clustering point cloud using the time series accumulation data of the clustering point cloud, and the velocity vector of the clustering point cloud that can be tracked is calculated. For example, the centroid position (three-dimensional position) at each time of the clustering point group for the same object is calculated, and the velocity vector is calculated from the centroid position at each time. The speed vector of the clustering point group includes information on the moving direction and moving speed of the clustering point group.

  In the point cloud movement information determination process, for each clustering point cloud, the movement information (velocity vector, centroid position information, etc.) of the clustering point cloud calculated by the point cloud tracking process is time-sequential (the movement of the clustering point cloud). Accumulate time). In the point cloud movement information determination process, for each clustering point group, the clustering point group continues to move stably for a certain period of time based on the accumulated movement information of the clustering point group (continues movement with little change). Or not). In this determination, for example, it is determined whether the time change amount of the moving speed is less than the threshold value and the time change amount in the moving direction is less than the threshold value for a certain time, or whether the time change amount of the distance between the center of gravity positions is less than the threshold value. judge. These threshold values and fixed time are set by experiments or the like.

  Trees and banners that fluctuate due to the wind may fluctuate periodically, but their moving speed and direction change randomly at short time intervals. On the other hand, a pedestrian continuously moves to a predetermined direction at a predetermined speed to some extent, and changes in the moving direction and moving speed with time are small. Therefore, in the point cloud movement information determination process, the clustering point group determined not to continue moving stably for a certain time is determined as a stationary object having fluctuations and excluded from the clustering point group to be processed.

  In the variation amount calculation process, the movement direction of the clustering point group (the velocity vector calculated by the point group tracking process) is determined for each clustering point group that is determined to have continued stable movement for a certain period of time in the point group movement information determination process. The amount of variation with respect to (direction) is calculated. Examples of the variation amount with respect to the moving direction include the length of the clustering point group in the moving direction and the variance value of the moving direction of the clustering point group.

  In the variation amount normalization processing, for each clustering point group, the variation amount calculated in the variation amount calculation processing is normalized according to the moving speed of the clustering point group (speed of the velocity vector calculated in the point group tracking processing). . As normalization according to the movement speed of the variation amount, for example, the variation amount is divided by the movement speed.

  For pedestrians, the stride is small when walking (when the speed is slow), so the amount of variation in the direction of movement is small, and the stride is large when running (when the speed is fast). The amount of variation with respect to the direction increases. That is, in the case of a pedestrian, the speed of the clustering point group and the amount of variation with respect to the moving direction have a correlation specific to the pedestrian, and even if the speed changes, the change in normalized variation is small. As a result, by performing periodicity determination using the normalized variation amount, even when the pedestrian speed changes, the pedestrian can be accurately identified. However, in the case of a bicycle, the variation in the amount of variation with respect to the direction of movement may have periodicity depending on the movement of the person's legs, but the amount of variation in the direction of movement does not change even when the speed is slow or fast. Therefore, in the case of a bicycle, when the speed changes, the normalized variation also changes. Further, even if there is a case where the variation of the variation amount with time has periodicity due to a measurement error of the laser radar 2, there is no correlation between the variation amount with respect to the speed and the moving direction.

  In the periodicity determination process, for each clustering point group, the normalized variation amount calculated in the variation amount normalization process is time-sequential for a predetermined time (a sufficient amount of time to determine the periodicity of the variation amount in the moving direction). )accumulate. In the periodicity determination processing, the periodicity of the temporal variation of the normalized variation amount is calculated for each clustering point group using the accumulated normalized variation amount accumulated data. In calculating the periodicity of the temporal variation of the normalized variation amount, for example, the difference between the normalized variation amount calculated at an arbitrary time t and the normalized variation amount calculated at the previous time t−1 is calculated. Further, in the periodicity determination process, for each clustering point group, it is determined whether the temporal variation of the normalized variation amount is periodic based on the periodicity of the temporal variation of the normalized variation amount. When there is periodicity, in the periodicity determination process, the attribute of the clustering point group is identified as a pedestrian, and information such as three-dimensional position information, moving direction, and moving speed is acquired from the clustering point group. On the other hand, when there is no periodicity, in the periodicity determination process, the attribute of the clustering point group is identified as other than a pedestrian.

  For example, in the case of the example shown in FIG. 2C, the moving speeds at the times t1, t2, t3, and t4 are set to v at the same speed, and the normalized variation amounts are L31 / v, L32 / v, L33 / v, L34 / v. In this example, the difference in normalized variation is (L32−L31) / v between times t2 and t1, and the difference in normalized variation is (L33−L32) / v between times t3 and t2. The difference in normalized variation between times t4 and t3 is (L34−L33) / v. Since (L32-L31) / v is a positive value, (L33-L32) / v is a negative value, and (L34-L33) / v is a positive value, the absolute values of the values are almost the same. The periodic variation in the amount of variation will be periodic.

  With reference to FIG. 1, the operation | movement in the pedestrian detection apparatus 1 is demonstrated. In particular, the processing of the ECU 3 will be described along the flowcharts of FIGS. FIG. 3 is a flowchart showing a flow of main processing of pedestrian detection according to the present embodiment. FIG. 4 is a flowchart showing a flow of point cloud movement information determination processing according to the present embodiment. FIG. 5 is a flowchart showing a flow of variation amount normalization processing according to the present embodiment. FIG. 6 is a flowchart showing a flow of periodicity determination processing according to the present embodiment.

  The laser radar 2 scans in a horizontal direction and / or a vertical direction at regular time intervals, emits laser light at each scanning angle, and receives reflected light. Then, the laser radar 2 transmits information about each detected point that can be received as a radar signal to the ECU 3.

  Every time a radar signal at each time is received, the ECU 3 calculates a relative distance to the detection point from information on the detection point included in the radar signal for each detection point, and calculates a three-dimensional position (S1). Here, a three-dimensional point group D1 composed of all detection points detected by the laser radar 2 at the current time is obtained.

  The ECU 3 estimates detection points for the road plane from detection points included in the three-dimensional point group D1 (S2). Here, a road plane point group D2 including detection points for the road plane detected at the current time is obtained. Furthermore, the ECU 3 excludes detection points included in the road plane point group D2 from detection points included in the three-dimensional point group D1, and separates detection points for the three-dimensional object (S3). Here, a solid object point group D3 including detection points for all the solid objects detected at the current time is obtained. Further, the ECU 3 clusters the detection points included in the three-dimensional object point group D3 (S4). Here, a clustering point group D4 composed of a plurality of detection points at the current time is obtained.

  The ECU 3 tracks the clustering point group D4 for each clustering point group D4 using the time series accumulation data of the clustering point group D4, and calculates the velocity vector of the clustering point group D4 (S5). Here, the velocity vector D5 for each clustering point group D4 at the current time is obtained.

  Next, the ECU 3 shifts to a point cloud movement information determination process (S6). When shifting to the point group movement information determination process, the ECU 3 accumulates movement information (speed vector D5, etc.) of the clustering point group D4 for a predetermined time in time series for each clustering point group D4 (S60). Here, movement information accumulation data D6 for each clustering point group D4 for a predetermined time until this time is obtained. Then, the ECU 3 determines, for each clustering point group D4, whether or not the clustering point group D4 is stably moving for a predetermined time based on the movement information accumulation data D6 (S61). When it is determined in S61 that the clustering point group D4 has not moved stably, the ECU 3 identifies the clustering point group D4 as a stationary object having fluctuations, and the clustering point group D4 is subsequently processed with respect to the clustering point group D4. Processing is not performed (S62).

  For each clustering point group D4 determined to be stably moving in S61, the ECU 3 calculates a variation amount with respect to the moving direction of the clustering point group D4 (the direction of the velocity vector D5) (S7). Here, a variation amount D7 for each clustering point group D4 at the current time is obtained.

  Next, the ECU 3 proceeds to a variation amount normalization process (S8). When shifting to the variation amount normalization process, the ECU 3 normalizes the variation amount D7 for each clustering point group D4 according to the moving speed of the clustering point group D4 (speed of the speed vector D5) (S80). Here, the normalized variation amount D8 for each clustering point group D4 at the current time is obtained.

  Next, the ECU 3 proceeds to a periodicity determination process (S9). When shifting to the periodicity determination process, the ECU 3 accumulates the normalized variation amount D8 for a predetermined time in time series for each clustering point group D4 (S90). Here, normalized variation accumulation data D9 for each clustering point group D4 for a predetermined time until the current time is obtained. Then, the ECU 3 calculates the periodicity of the temporal variation of the normalized variation amount for each clustering point group D4 using the normalized variation amount accumulation data D9 (S91). Further, the ECU 3 determines, for each clustering point group D4, whether or not the temporal variation of the normalized variation amount is periodic based on the periodicity of the temporal variation of the normalized variation amount (S92). When it is determined that there is periodicity in S92, the ECU 3 identifies the attribute of the clustering point group D4 as a pedestrian and acquires information such as three-dimensional position information from the clustering point group D4 (S93). On the other hand, if it is determined in S92 that there is no periodicity, the ECU 3 identifies the attribute of the clustering point group D4 as other than a pedestrian (S94).

  In ECU3, when a pedestrian can be identified by the process of this time, the information about the pedestrian is output to a driving assistance apparatus etc.

  According to this pedestrian detection device 1, a pedestrian can be identified with high accuracy by identifying a pedestrian based on whether or not there is periodicity in the variation in the amount of variation with respect to the moving direction of the clustering point group. . This can reduce erroneous detection of a clustering point group for an object having periodicity due to other than pedestrian movement. Of course, a clustering point group for an object having no periodicity is not erroneously detected. Furthermore, it is possible to identify a moving object other than a pedestrian such as a normal vehicle or a large vehicle, a motorcycle, a bicycle, or the like that exists in a general traffic environment with high accuracy. Therefore, a pedestrian can be detected with high accuracy for the entire general traffic environment.

  Moreover, according to the pedestrian detection apparatus 1, by using the normalized variation amount obtained by normalizing the variation amount of the clustering point group in accordance with the movement speed, the pedestrian can be obtained from the movement speed of the clustering point group and the variation amount with respect to the movement direction. Whether or not there is a specific correlation can be determined with high accuracy, and a pedestrian can be detected with higher accuracy. For example, even if the variation amount of the clustering point group varies periodically due to the measurement error of the laser radar 2, the pedestrian-specific correlation exists between the movement speed and the variation amount with respect to the movement direction for the clustering point group. Since the possibility is very low, it is possible to reduce erroneous detection of the clustering point group. In addition, in the case of other moving bodies (for example, bicycles) whose variation in time varies with the periodicity similar to that of pedestrians, even if the moving speed changes, the amount of variation in the moving direction does not change. There is no correlation, and erroneous detection of such a moving object can be reduced. Moreover, even when a pedestrian is walking or running, it can be accurately identified as a pedestrian regardless of the moving speed.

  Moreover, according to the pedestrian detection apparatus 1, it is possible to detect a pedestrian with higher accuracy by determining whether or not the clustering point group is moving stably for a certain period of time. For example, in the case of a stationary object with periodic motion (for example, a flag or tree that swings in the wind), its moving direction and speed change randomly in a short time, so such a stationary moving object is erroneously detected. Can be reduced.

  As mentioned above, although embodiment which concerns on this invention was described, this invention is implemented in various forms, without being limited to the said embodiment.

  For example, although applied to the pedestrian detection apparatus mounted in a vehicle in this Embodiment, it is applicable also to the pedestrian detection apparatus used other than a vehicle.

  In the present embodiment, the detection means is a laser radar, but other means such as a millimeter wave radar, an ultrasonic radar, and a stereo camera may be used. In the present embodiment, three-dimensional position information is acquired as detection point information, and each process is performed using the three-dimensional position information. However, detection point information includes two-dimensional position information (height information). None) is possible.

  In this embodiment, in order to extract a clustering point group from all detection points detected by the laser radar, one of the conventional methods using distance measurement processing, road plane estimation processing, three-dimensional object separation processing, and clustering processing is used. However, any method may be applied as a method for extracting the clustering point group. In this embodiment, clustering is used to group detection points into a plurality of detection point groups. However, grouping may be performed by other methods.

  Further, in this embodiment, in order to determine whether or not there is a pedestrian-specific correlation between the movement speed of the clustering point group and the fluctuation quantity with respect to the movement direction, normalization is performed by normalizing the fluctuation quantity according to the movement speed. Although the variation amount is used, a parameter other than the normalized variation amount may be obtained to determine whether or not there is a pedestrian-specific correlation between the movement speed and the variation amount with respect to the movement direction.

  In this embodiment, the point cloud movement information determination process and the variation amount normalization process are performed. However, either one process or both processes may not be performed.

  In addition, when identifying a pedestrian, attention is paid to the periodicity caused by the movement of the legs, and thus the lower body of the pedestrian is characteristic. Therefore, the clustering point group may be further divided into an upper point group and a lower point group based on the height information, and each process may be performed using only the lower point group. By performing such a vertical separation process, noise due to a measurement error by a laser radar or the like can be suppressed, and detection accuracy can be further improved.

  Further, the vehicle is clearly larger when comparing the size of the pedestrian and the vehicle, and the vehicle is clearly faster when comparing the speed of the pedestrian and the vehicle. Other mobile objects also differ from pedestrians in size and speed. Therefore, the clustering point cloud that has a clearly large size (size that cannot be clearly regarded as a pedestrian) is excluded, or the clustering point cloud that has a clearly fast speed (can be regarded as a pedestrian clearly). For example, it may be possible to narrow down the clustering point group of the pedestrian attribute and perform each processing using only the narrowed clustering point group. By performing such an attribute narrowing process, the processing load can be reduced (the processing time is shortened), and the data amount of the variation accumulation data can be reduced (the memory amount is reduced).

  DESCRIPTION OF SYMBOLS 1 ... Pedestrian detection apparatus, 2 ... Laser radar, 3 ... ECU.

Claims (3)

The detection points detected by the detection means are grouped as a plurality of detection point groups, and a pedestrian detection device that detects pedestrians from these detection point groups,
A moving direction calculating means for calculating the moving direction of the detected point group for each detected point group;
Distribution information calculation means for calculating distribution information of the detection point group with respect to the movement direction calculated by the movement direction calculation means for each detection point group;
For each detection point group, it is determined whether the temporal change in the distribution information of the detection point group with respect to the moving direction calculated by the distribution information calculation means is periodic, and the detection point group having periodicity is determined as a pedestrian Determination means to perform,
A pedestrian detection device comprising:   2. The walking according to claim 1, wherein the determination unit determines the detection point group as a pedestrian when there is a correlation specific to the pedestrian in the distribution information with respect to the movement speed and the movement direction of the detection point group. Person detection device.   The determination means accumulates the movement information of the detection point group for each detection point group, and determines that the detection point group is moving stably for a predetermined time based on the accumulated movement information of the detection point group. The pedestrian detection device according to claim 1 or 2, wherein the detection point group is determined to be a pedestrian.

Images