JP2010205042A - Pedestrian detection device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a pedestrian detection device for accurately detecting a pedestrian. <P>SOLUTION: A three-dimensional location information acquisition part 11 acquires the three-dimensional location information of the periphery of a vehicle in a time-sequential order. An analyzing part 12 expresses the three-dimensional location information at each time as a point group on a plane on which the vehicle is located, and analyzes the periodicity of the distribution change of the point group following the time series. When it is determined that there exists periodicity by the analyzing part 12, the point group is detected as a pedestrian. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a pedestrian detection technique in a pedestrian detection device mounted on a vehicle.

  As a pedestrian detection apparatus as described above, a technique for detecting a pedestrian by detecting a human-specific shape pattern from an image taken using an infrared camera is known (see Patent Document 1).

  In this technique, a candidate area for a pedestrian determined by preprocessing is cut out from an image captured by an infrared camera. Whether or not the candidate area is a pedestrian is determined based on the degree of similarity between the infrared intensity pattern of the extracted candidate area and the infrared pattern (temperature pattern and shape pattern) peculiar to humans.

JP-A-2005-157765 (paragraph numbers 0007-0011)

  In the technique of Patent Document 1, since an infrared camera is used, a desirable infrared image cannot be obtained when a heat source other than a human is present. That is, it cannot be used during the day. Also, since the pattern of infrared intensity varies greatly depending on the situation of the pedestrian's clothes and the like, it is difficult to prepare a pattern that covers all situations.

  The objective of this invention is providing the pedestrian detection apparatus which detects a pedestrian exactly.

  In order to solve the above-described problem, a pedestrian detection device according to the present invention includes a three-dimensional position information acquisition unit that acquires three-dimensional position information around a vehicle in chronological order, and the vehicle positions the three-dimensional position information. Express as a point cloud on the plane, and analyze the periodicity of the distribution change of the point cloud along the time series, and walk the point cloud when the analysis unit determines that there is periodicity Detect as a person.

  With this configuration, the three-dimensional position information around the vehicle acquired by the three-dimensional position information acquisition unit is converted into a point cloud on the plane on which the vehicle is located, and the analysis unit changes the distribution of the point cloud in time series. Analyze the periodicity of. When the movement of a pedestrian is projected on the plane, the movement of arms and legs has periodicity. The said structure is for analyzing this periodicity, and the point cloud determined with the periodicity by the analysis part is detected as a pedestrian.

  Since the movement of the arm may not be able to analyze the periodicity due to the influence of the luggage or the like, it is preferable to analyze the periodicity by the movement of the legs. Therefore, it is desirable to acquire only the height corresponding to the vicinity of the human leg from the three-dimensional position information. By limiting the range in which the three-dimensional position information is acquired in this way, the acquisition time of the three-dimensional position information can be reduced, and other parts of the pedestrian become noise, reducing the accuracy of the periodicity analysis. This can be prevented.

It is a functional block diagram of the pedestrian detection device of the present invention. It is a flowchart showing the flow of a process of the pedestrian detection apparatus of this invention. It is a figure showing a motion of a pedestrian's leg. It is a figure which shows the distribution change of the point cloud showing the leg of a pedestrian. It is a figure which shows the distribution change of the point cloud showing the leg of a pedestrian. It is a figure which shows the distribution change of the point cloud showing the leg of a pedestrian. It is a flowchart showing the flow of the process of the periodicity analysis in this invention.

  Embodiments of a pedestrian detection device of the present invention will be described below with reference to the drawings. FIG. 1 is a functional block diagram showing each functional part of the pedestrian detection device of the present invention, and FIG. 2 is a flowchart showing the flow of pedestrian detection processing. As shown in the figure, the pedestrian detection device is mounted on a vehicle, and in order to detect a pedestrian in the vehicle traveling direction, a three-dimensional position information acquisition unit 11 that acquires distance data ahead of the vehicle in the traveling direction. An arithmetic processing unit A that detects a pedestrian in front of the vehicle by analyzing the distance data acquired by the position information acquisition unit 11 is provided.

  In the present embodiment, the three-dimensional position information acquisition unit 11 measures the distance based on a signal from the infrared laser radar L. As described above, since the periodicity of the movement of the pedestrian's leg is easy to analyze, the infrared laser radar L is installed at a position where the three-dimensional position information of the pedestrian's leg can be measured. For example, it can be installed near the bumper of a vehicle. Naturally, by changing the installation direction of the infrared laser radar L, a pedestrian in an arbitrary direction can be detected. In the following description, the width direction of the vehicle is the x axis, the height direction of the vehicle is the y axis, and the traveling direction of the vehicle is the z axis. Therefore, the plane on which the vehicle is located is the xz plane.

The infrared laser radar L is configured to be able to scan in the x-axis direction with a polygon mirror or the like, and to project four lines in the y-axis direction at intervals of 0.8 °. Further, the laser scan is executed at predetermined time intervals. The three-dimensional position information acquisition unit 11 acquires three-dimensional position information ahead of the vehicle every predetermined time based on the signal from the infrared laser radar L configured as described above (# 11). In the present embodiment, the predetermined time is 100 ms. This three-dimensional position information is stored in the RAM as a point group {P _ti } on the xz plane. Note that t represents the time when the distance data was acquired, and i represents the number of the point data. At this time, the three-dimensional position information acquisition unit 11 notifies the analysis unit 12 that the three-dimensional position information has been acquired.

In the analysis unit 12, the clustering unit 12a acquires the point cloud data { _Pti } from the RAM, and performs clustering on the xz plane (# 12). This is because when a pedestrian exists, the point group representing the pedestrian is considered to form a cluster having a certain shape on the xz plane. As a clustering method, a known method such as a K-Mean method or vector quantization can be used. By this clustering, the point group {P _ti } is classified into a plurality of clusters {C _j }, and each cluster C _j belongs to a point group that is close in the xz plane.

  Since there are various objects in addition to the pedestrian in front of the vehicle, it is necessary to select a cluster composed of point groups of the pedestrian's legs in order to accurately detect the pedestrian. Therefore, the clustering unit 12a extracts only clusters (hereinafter referred to as pedestrian candidate clusters) that may be composed of point groups of the pedestrian's legs based on a predetermined condition (# 13). In this embodiment, a cluster in which the distribution of point groups belonging to the cluster is within a circle having a diameter of 1 m is extracted as a pedestrian candidate cluster. Specifically, if the difference between the minimum x coordinate and the maximum x coordinate is within 1 m and the difference between the minimum z coordinate and the maximum z coordinate is within 1 m in the point group in the cluster, the cluster is selected as a pedestrian candidate. A cluster. However, if the number of points belonging to the cluster is extremely small even if this condition is satisfied, the accuracy of the subsequent verification cannot be obtained sufficiently and is excluded from the pedestrian candidate cluster.

Next, the pedestrian candidate cluster is verified by the periodicity analysis unit 12b of the analysis unit 12. That is, it is verified whether or not the point group belonging to the pedestrian candidate cluster is a pedestrian. 3 (a) to 3 (c) are images of pedestrian legs crossing the front of the vehicle, and FIGS. 4 (a) to 4 (c) are point groups (one pedestrian candidate cluster) representing those legs. It is the figure which plotted the point group which belongs to xz plane. FIG. 4 shows the xz plane divided into four quadrants with the median of the point group belonging to the cluster as the origin. In addition, the median in this embodiment means the average value of the maximum value and the minimum value. That is, if the point group belonging to the pedestrian candidate cluster is {P _t1 (x ₁ , z ₁ ),..., P _tn (x _n , z _n )}, the median is ((max ({x _i } ) −min ({x _i })), (max ({z _i }) − min ({z _i }))).

  The leg of the pedestrian repeats the movement of (a) → (b) → (c) → (b) → (a) in FIG. Along with this, the change of the point group also repeats (a) → (b) → (c) → (b) → (a) in FIG. As is clear from these, the change in the quadrant where the point cloud is distributed with the walking motion has periodicity.

  This change in periodicity can be seen from another point of view. Since the point group belonging to the cluster is plotted on the xz plane as described above, each point is represented by the distance (radial radius) r from the median value (origin) and the angle θ formed by the radial radius and the x axis. (See FIG. 4A). At this time, each point is plotted on a plane (hereinafter referred to as the θr plane) having the angle θ on the horizontal axis and the radius r on the vertical axis, as shown in FIG. FIGS. 5A to 5C are diagrams in which the points of FIGS. 4A to 4C are plotted on the θr plane. Here, the θr plane is divided into regions of 90 ° (the divided regions are referred to as the first quadrant to the fourth quadrant in order of angle). In FIG. 5A, the points are concentrated in the second and fourth quadrants, and in FIG. 5C, the points are concentrated in the first and third quadrants. Moreover, in (b), it distributes to each quadrant. Therefore, the change in the distribution of points on the θr plane accompanying the walking motion also has the periodicity as described above.

Also, periodicity can be seen in the distribution of the point cloud distribution. For example, the degree of dispersion of the point cloud distribution can be obtained by the equation (1), and the average value of distances from the median value m in the xz plane of the point cloud belonging to the pedestrian candidate cluster (hereinafter simply referred to as the degree of dispersion σ). Can be expressed as Here, distance () is a function for obtaining the distance between two points. At this time, when the variance σ _t of the point group belonging to the pedestrian candidate cluster at time _t is obtained and plotted on a plane with time t on the horizontal axis and the variance σ on the vertical axis, it is close to a sine curve as shown in FIG. Draw a curve. Therefore, in this case, periodicity can be obtained analytically. Therefore, in this embodiment, whether or not the pedestrian candidate cluster is a pedestrian is verified by performing periodicity analysis focusing on the distribution of the distribution of the point cloud.

  The periodicity analysis unit 12b analyzes the periodicity of the distribution change of the point cloud representing the pedestrian's legs as described above (# 14). Specifically, the process of the flowchart of FIG. 7 is performed.

  First, in order to analyze periodicity, data of a certain time (for example, 10 seconds) is necessary. Therefore, it is determined whether or not point cloud data for a predetermined time is stored in the RAM (# 21). If data for a predetermined time is not acquired, the process is terminated and the process returns to the main routine (#). 21 No branch).

  On the other hand, when the point cloud for the required time is stored in the RAM (Yes branch of # 21), the periodicity analysis is started. Here, a point cloud from time t to time t + n is recorded in the RAM, and it is assumed that there is only one pedestrian candidate cluster at each time. When there are a plurality of pedestrian candidate clusters, it is necessary to take correspondence of the pedestrian candidate clusters in the time direction. In this case, if the distance between the median value m on the xz plane of the pedestrian candidate cluster at time t and the median value m on the xz plane of the pedestrian candidate cluster at time t + 1 are compared, it can be easily determined in the time direction. Tracking can be done.

First, the dispersion degree σ of the point group at each time in the time interval [t, t + n] is obtained using the expression (1) (# 22), and the time when the dispersion degree σ _t becomes a maximum (hereinafter referred to as a maximum time). {T _i } and a minimum time (hereinafter referred to as a minimum time) {B _i } are obtained (# 23).

Due to the maximum time T _i and the minimum time B _i , the time interval [t, t + n] is divided into a plurality of time intervals [t, T _i ], [T _i , B _i ], [B _i , T _{i + 1} ]. , [T _{i + 1} , B _{i + 1} ],. Hereinafter, the divided time section is referred to as a divided section.

Next, one division section is selected (# 24), and the sum S of the variance differences between the times is obtained in the selected division section (# 25). That is, if the start point of the time interval is t and the end point is t + k, the sum S can be obtained by equation (2).

  At this time, if S is outside the predetermined threshold range (No branch of # 26), it is determined that the person is not a pedestrian and the process is interrupted, and the process returns to the main routine. Since the total sum S is considered to correspond to the pedestrian's stride, it is desirable that the predetermined threshold range is set to a range that can be taken by the pedestrian's stride.

  On the other hand, if the sum S is within the predetermined threshold range (Yes branch of # 26), the processing is continued assuming that there is a possibility of a pedestrian. If there is an unprocessed divided section (Yes branch of # 27), the process proceeds to # 24, and if there is no unprocessed divided section (No branch of # 27), the process proceeds to # 28.

  With the above processing, the point group changes in distribution about the pedestrian's stride. Therefore, in the next process, the periodicity of the change is verified (# 28 to 31). For example, the periodicity can be determined based on the variation in the length of each divided section. The length of each divided section represents the time for one step of the pedestrian. Therefore, in an ideal pedestrian, the length of each divided section is constant. In this embodiment, this characteristic is used. That is, the variance of the time length of each divided section is obtained (# 28), and if the variance is smaller than a predetermined threshold (Yes branch of # 29), it is determined that there is periodicity (# 30), and is equal to or greater than the predetermined threshold If so (No branch at # 29), it is determined that there is no periodicity (# 31). As described above, since the length of the divided section represents the time of one step of the pedestrian, that is, the walking speed (pitch), the range of the walking pitch that can be taken by the pedestrian is set and set, and the length of the divided section is set. May be determined not to be a pedestrian when the distance is outside the walking pitch range. These verification results are notified to the control unit 13.

  Note that the periodicity analysis method of the periodicity analysis unit 12b is not limited to the above-described method, and a known method such as Fourier analysis may be used.

  The control unit 13 that has received the periodic analysis result warns the driver via the speaker S because there is a high possibility that there is a pedestrian in the case of the result of periodicity (Yes branch of # 15). A sound or warning sound is emitted (# 16). On the other hand, if the result indicates that there is no periodicity (No branch of # 15), the above-described processing is repeated.

  In the above-described embodiment, the method for detecting a pedestrian crossing the front of the vehicle has been described. However, a pedestrian such as a pedestrian who walks in parallel with the traveling direction of the vehicle can be detected in the same manner.

  It can be used to detect an obstacle in the traveling direction of the vehicle.

11: Three-dimensional position information acquisition unit 12: Analysis unit

Claims (1)

A three-dimensional position information acquisition unit for acquiring three-dimensional position information around the vehicle in chronological order;
An analysis unit that expresses the three-dimensional position information as a point cloud on a plane on which the vehicle is located, and analyzes the periodicity of the distribution change of the point cloud along the time series;
A pedestrian detection apparatus that detects the point cloud as a pedestrian when the analysis unit determines that there is periodicity.

Images