JP4631036B2 - Passer-by behavior analysis device, passer-by behavior analysis method, and program thereof - Google Patents

Description

  The present invention relates to a passer-behavior analysis device, a passer-behavior analysis method, and a program for analyzing a passer-by behavior in a space where a passer-by moves and stays at a railway station or the like.

Passengers using the station can flow and move smoothly is an important requirement for the comfort of the station, but passengers and transfer passengers are complicated, stoppages and shops to check the signs and entrances and exits There are many factors that impede the smooth flow, such as shopping at a store or staying at a meeting. This causes problems such as a decrease in walking speed due to passenger collisions and congestion.
Conventional passenger flow surveys, which have been conducted as basic information collection for studying improvement works that improve the comfort of the station environment, are mainly to grasp the number of people passing through the section, and manual surveys are the mainstream. Therefore, it was difficult to comprehensively analyze the speed, direction, and density of passengers moving across the station.

  In order to solve such a problem, a camera of a visible image is installed at a road intersection, and an image in a monitoring area of a predetermined range including a pedestrian crossing is captured by this camera, and a pedestrian at a road intersection is captured. By examining the situation and the number of people, a technique is known that can more efficiently control the signal for the vehicle and the signal for the pedestrian, and as a result, improve the safety of the pedestrian (see, for example, Patent Document 1). ).

Here, since the device shown in Patent Document 1 uses visible video images, tracking processing may be difficult when passers-by interlaces, and accurate tracking processing of passers-by is performed. There is a problem that it is difficult to do. In addition, in order to avoid this problem, the installation position and the installation angle of the camera are greatly restricted, the imaging range becomes narrow, and there are problems that the calculation processing load increases when there are many passersby. And the technique which can solve such a problem and can analyze a passerby's action using a distance picture is disclosed (for example, refer to patent documents 2).
JP 11-275562 A JP 2005-346617 A

  However, in the technique of Patent Document 2, the accuracy of the passerby tracking process is not necessarily sufficient, and it is desired to increase the accuracy of the process.

  Accordingly, an object of the present invention is to provide a passer-by behavior analysis apparatus that can improve the accuracy of passer-by tracking processing when a large number of people are present in a relatively large place.

  In order to achieve the above object, the present invention uses distance data receiving means for receiving distance data indicating the distance of a passer's foot from a laser sensor, and a plurality of the distance data acquired by a plurality of the laser sensors. From the clustering result, sensing image data generating means for generating sensing image data indicating the detection position of the laser sensor on the spatial plane at the same time, clustering means for clustering a plurality of the detection positions in the sensing image data, and From the passer identification means for specifying the cluster corresponding to the foot of one passer-by, and the position of the passer's foot, the walking direction, the stride, the walking cycle, and the walking phase from the identified passer-by cluster Parameter determining means for determining the parameter, and the parameter and the particle filter algorithm The locus analysis means for analyzing the trajectory of passers by using prism, a passerby behavioral analysis apparatus comprising: a.

  Further, the present invention is characterized in that the above-mentioned passerby behavior analysis apparatus uses a Mean-shift Clustering algorithm in the clustering.

  Further, the present invention is characterized in that, in the above-described passer-behavior analysis device, a cluster that is divided or integrated by the cluster of the clustering results is identified as a cluster corresponding to the leg of the one passer-by.

In the present invention, the trajectory analysis processing unit of the above-mentioned passer-by behavior analysis apparatus , from the acceleration of the left foot and the right foot of the passer-by identified in the clustering result cluster in the processing of the Particle filter algorithm , The walking phase of the parameter is updated, the movement distance of the passerby is updated by the walking phase, and the foot point of the passerby of the parameter is updated from the updated walking phase, the movement distance, and the walking direction. It is characterized by predicting the position of .

  The present invention is also a passer-by behavior analysis method in a passer-by behavior analysis apparatus, which is obtained by a distance data receiving process for receiving distance data indicating a distance from a laser sensor of a passer's foot, and a plurality of the laser sensors. A plurality of the detected distance data, and sensing image data generation processing for generating sensing image data indicating the detection position of the laser sensor on the space plane at the same time, and clustering the plurality of detection positions in the sensing image data Clustering processing, passer identification processing for identifying a cluster corresponding to a passer's foot from the clustering result, and position of the passer's foot, walking direction, stride from the identified passer's cluster A parameter determination process for determining a parameter consisting of a walking cycle and a walking phase; A passerby behavioral analysis method characterized by having a trajectory analysis process for analyzing the trajectory of passers by using the serial parameters and Particle filter algorithm.

  Further, the present invention provides a computer of a passer-by behavior analysis device, distance data receiving means for receiving distance data indicating a distance of a passer's foot from a laser sensor, and a plurality of the distance data acquired by the plurality of laser sensors. The sensing image data generating means for generating sensing image data indicating the detection position of the laser sensor on the spatial plane at the same time, clustering means for clustering a plurality of the detection positions in the sensing image data, and one person from the clustering result A passer specifying means for specifying a cluster corresponding to a passer's foot, a parameter comprising the position of the passer's foot, the walking direction, the stride, the walking cycle, and the walking phase from the specified passer's cluster Parameter determining means for determining the parameter and the particle filter Is a program for causing to execute a trajectory analysis means, for analyzing the trajectory of passers using Gorizumu.

  ADVANTAGE OF THE INVENTION According to this invention, even when a passerby is crowded, the analysis method of the locus | trajectory of each passerby can be provided accurately.

Hereinafter, a passer-by behavior analysis apparatus according to an embodiment of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram showing a configuration of a passer-by behavior analysis system according to the embodiment.
In this figure, reference numeral 1 denotes a passer-by behavior analysis device that executes a behavior analysis process of a passer-by 7 in a predetermined space 6. Reference numeral 2 denotes a laser sensor that measures the distance to an object such as a passerby. The laser sensor 2 scans a horizontal plane at a height of about 10 cm from the floor, and is a horizontal section corresponding to the height of the laser projection, and distance data in an orthogonal coordinate system that includes both stationary and moving objects. Can be obtained. Reference numeral 3 denotes a client PC (personal computer) provided for each laser sensor 2 in order to control each laser sensor 2. Reference numeral 4 denotes a network that performs information communication between each client PC 3 and the passer-by behavior analysis apparatus 1. The measurement data stored in the client PC 3 is transferred to the passer-by behavior analysis apparatus 1, and is sent from the server 1. It is used to send instruction information to the client PC 3.

Next, the configuration of the passer-by behavior analysis apparatus 1 will be described with reference to FIG.
FIG. 2 is a block diagram showing a configuration of the passer-by behavior analysis apparatus 1 shown in FIG.
In this figure, reference numeral 11 denotes a communication processing unit that receives distance data from each client PC 3. Reference numeral 12 denotes a sensing data storage unit for storing distance data received by the data receiving unit 11. Reference numeral 13 denotes a parameter information storage unit that stores various parameters for performing an analysis process of a passerby action based on the distance data. Further, reference numeral 14 denotes a time-synchronizing process that extracts a foot portion of the passerby 7 from sensing image data integrated in one coordinate system, and performs a tracking process for tracking the passerby 7. Is a trajectory calculation unit that calculates a trajectory of a passerby who has passed within a predetermined time. Reference numeral 15 denotes a trajectory data storage unit that stores the pass trajectory data of each passer-by obtained by the trajectory calculation unit 14. Reference numeral 16 denotes a trajectory output unit that outputs a trajectory image based on the traffic trajectory data. Reference numeral 17 denotes a display unit configured by a display.

Next, an operation of extracting and tracking the passerby 7 will be described.
First, each client PC 3 transfers the distance data obtained from the laser sensor 2 to the passer-by behavior analysis apparatus 1. In response to this, the communication processing unit 11 records the distance data received from each client PC 3 in the sensing data storage unit 12.

  Next, the trajectory calculation unit 14 reads a plurality of distance data stored in the sensing data storage unit 12 to obtain temporal synchronization and generate sensing image data integrated into one coordinate system. When the sensing image data is generated, the trajectory calculation unit 14 starts tracking processing.

FIG. 3 is a diagram showing an outline of sensing image data.
As shown in this figure, sensing image data is generated by receiving distance data specified by laser light reflected from a leg of a passerby from a plurality of laser sensors 2, and outputting the sensing image data. Displays a laser sensor, a foot point of a passerby, and a background (such as a wall or a pillar). In this embodiment, the laser sensor 2 transmits a laser beam at a height of 10 cm from the ground surface and detects the distance near the passer's heel by its reflection, but this is the height near the waist. This is because when many people cross each other, the degree of congestion of the measurement object in the space increases, making measurement difficult.

FIG. 4 is a first diagram showing an outline of the walking model.
Next, tracking processing by the locus calculation unit will be described.
First, the trajectory of the passer's foot is periodic. That is, it is considered that the data obtained by the laser sensor 2 can acquire periodic and net-like data as shown in FIG. 4 from two legs of a passerby. Therefore, the walking model according to the present embodiment, using a Particle filter algorithm takes a method to predict the passers state. As a premise, (1) the pedestrian is walking in 2nd speed, and (2) the passerby does not attach any accessories near the foot so that the measurement of the passer's foot is not ambiguous. .

FIG. 5 is a second diagram showing an outline of the walking model.
When a passerby is walking, a typical foot swing mode can usually be seen. FIG. 5 shows the typical foot swing mode. As shown in this figure, the two feet of the passerby are alternately swung as roles of the two feet during walking. Then, as shown in FIG. 5, first, when a pedestrian thinks to walk from the right foot, the swing of the right foot is accelerated forward. And while the right foot is swinging, the force is gradually reduced and the swing of the right foot is stopped. Similarly, when the right foot stops, it swings with the acceleration of the left foot, gradually reducing the force, and the left foot stops. And the relationship of acceleration, speed, and moving distance as shown in FIG. 5 is derived. In FIG. 5, the broken line represents the right foot and the solid line represents the left foot. The acceleration of the right foot can be expressed as the following equation (1).

Here, in Expression (1), A is the maximum acceleration. T represents a walking cycle (one step for the left foot and one step for the right foot as one cycle). Therefore, the speed v _R (t) and the moving distance S _R (t) of the right foot can be expressed by Expression (2) and Expression (3).

Similarly, the movement distance S _L (t) of the left foot can be expressed by Expression (4).

Therefore, the distance S _d (t) between the two feet is obtained from the equations (3) and (4):

It can be expressed as. Also, the distance S _d (t) between the two feet is expressed as in the equation (6) as a cosine pattern S _d ^{^} (t) as shown in FIG. Can do.

In Equation (6), S indicates the stride and γ indicates the phase. In addition, it turns out that the pattern shown by S _R (t) and S _L (t) in FIG. 11 is similar to the periodic and string-like pattern shown in FIG. Further, the travel distance of the passerby in the walking phase can be calculated by the equation (6).

FIG. 6 is a diagram showing a clustering result.
Here, the trajectory calculation unit 14 clusters the point data of a plurality of passersby as shown in FIG. 6 using the sensing image data and the Mean-shift Clustering algorithm. In FIG. 6, the result of clustering marks up the feet of passersby with circles. The Mean-shift Clustering algorithm is described in non-patent literature <D. Comaniniciu and P. Meer, “Ditribution free decomposition of multivariate data”, Pattern Analysis and Applications, vol2, pp.22-30, 1999>. The method was used.

  Here, in the result of FIG. 6, clustering is usually performed for each foot, but when both feet of one passerby are close, these feet are clustered as one foot. In addition, the feet of different passersby may be clustered as one foot. This is because the Mean-shift Clustering algorithm automatically limits the cluster size.

FIG. 7 is a diagram showing an outline of integration and division of a plurality of laser points.
As shown in FIG. 7, the cluster of points obtained by clustering was one cluster in the scene of t−1 seconds, but may be clustered as two clusters in the scene of t seconds. Is a division. On the contrary, in the scene of t-1 seconds, there are two clusters, but in the scene of t seconds, there are cases where the clusters are clustered as one cluster, which is integration. In either case, they can be identified as both feet of a single passerby.

Here, in each successive frame of the sensing image data, first, the shape and size of each cluster and the moving direction are calculated based on the clustering result. The locus calculating section 14, for each cluster C _i ^t at time t, calculates the integration or split between adjacent clusters from the point of time t-1. Thereby, the trajectory calculation unit 14 determines which cluster pair or one according to the clustering result corresponds to one passer-by.

  Next, the trajectory calculation unit 14 starts a passerby tracking process. Here, the technique of the particle filter algorithm is effective for tracking a plurality of passers-by. In the passer-by tracking process using the particle filter algorithm, the following eight parameters are used.

  Lx and Ly indicate the position of the left foot, and Rx and Ry indicate the position of the right foot. S represents the stride, T represents the walking cycle, α represents the walking direction, and γ represents the phase. Then, the trajectory calculation unit 14 determines all passers-by at time t.

And the probability density function

Is calculated. Here, in Expression (8), Z ^{1: t} indicates all observation data (distance data from the laser sensor) until the time t is reached. Z ^t indicates observation data (distance data from the laser sensor) at time t. Further, in the equation (9), M indicates the number of samples, theta _{k, s} ^t denotes the s-th sample of the k-th object. The π _{k, s} ^t shows the corresponding weight.

Since the particle filter algorithm is an inductive function that repeats prediction and update, first, in the prediction step, a random number rε [0, 1] is generated, and the random number is used to generate time t−1 from all samples. Select one sample θ _k, ^st at. Regarding the processing of the particle filter algorithm, non-patent literature <M. Isard and A. Blake, “CONDENSATON-conditional density propagation for visual tracking”, Int. J. Computer Vision, vol. 29, no. 1, pp. 5-28 1998.> is used. Then, the following calculation is performed.

Here, Δt is the time from time t−1 to time t. Further, w _s , w _T , w _α , w _γ , w _r1 , and w _r2 are distributed random noises. In the prediction step, first, the walking phase γ and the movement distance md are updated by Expression (6). Next, based on the walking phase, it is determined whether the right foot or the left foot is moving, and a different dynamic range (r _L or r _R ) is assigned. Then, the position of both feet is predicted by the movement and the walking direction α. Finally, the stride S, the walking cycle T, and the walking direction α are predicted.

In addition, the update step is performed for each result detected by clustering. At this time, the weight (π _{k, s} ^t ) of all samples at time t is calculated for the _kth object. Observation target

The weight set is updated by the following formula.

As a result, p _L and p _R calculated indicate the positions of the left foot and the right foot. Further, p _B denotes the relative distance of the left foot and the right foot. D0 indicates the distance in the walking direction between the right foot and the left foot. Further, r0 is a reference value calculated using Equation (6).

  The values calculated by the equations (19) to (21) are temporarily stored from the locus calculation unit 14 to the locus data storage unit 15 at every calculation timing, and the locus output unit 16 performs the calculation at each calculation timing. A traffic trajectory image is output to the display unit 17 using the data.

FIG. 8 is a diagram showing a processing flow of the passer-by behavior analysis apparatus.
FIG. 9 is a diagram showing an outline of walking direction calculation.
As described above, the trajectory calculation unit 14 generates sensing image data based on the distance data received from each client terminal (step S1). Then, the laser points in the sensing image data are clustered using the Mean-shift Clustering algorithm (Step S2). Further, from the clustering result, a cluster in which two of the clusters at time t-1 and time t are integrated into one, or a cluster into which one is divided into two is identified as a single passer-by cluster (step S3). . Then, the direction shown in FIG. 9 is calculated from the position (Lx, Ly, Rx, Ry) of the point indicating the leg of one passerby and determined as the walking direction α, and the stride S, the walking cycle T, the walking Various parameters of the phase γ are determined (step S4). Then, using the various parameters, the tracking process is performed for all passers-by using the particle filter algorithm (step S5).

FIG. 10 is a diagram showing a traffic trajectory image.
As shown in this figure, the passer-by analysis apparatus 1 displays the trajectory of the pass at time t on the display unit 17. According to the process of this passer-by analysis apparatus 1, even when the passers-by is crowded, the trajectory of each passer-by can be analyzed with high accuracy. Then, using the processing of the above-described passer analysis device, tracking processing was performed on June 14, 2006, at Osaki Station (60 m × 20 m) in JR East Japan from 07:00 am to 07:05 am However, of all 387 passers-by, 367 passers were successfully tracked, resulting in a success rate of 94.8%. On June 14, 2006, the tracking process was performed at Osaki Station (60m x 20m) in JR East Japan from 08:00 am to 08:10 am, and about 80 out of 2608 passersby. % Passer-by tracking was successful.

  In addition, the above-mentioned passer-by analysis apparatus has a computer system inside. The process described above is stored in a computer-readable recording medium in the form of a program, and the above process is performed by the computer reading and executing this program. Here, the computer-readable recording medium means a magnetic disk, a magneto-optical disk, a CD-ROM, a DVD-ROM, a semiconductor memory, or the like. Alternatively, the computer program may be distributed to the computer via a communication line, and the computer that has received the distribution may execute the program.

  The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, and what is called a difference file (difference program) may be sufficient.

It is a block diagram which shows the structure of a passer-by action analysis system. It is a block diagram which shows the structure of a passer-by action analysis apparatus. It is a figure which shows the outline | summary of sensing image data. It is a 1st figure which shows the outline | summary of a walking model. It is a 2nd figure which shows the outline | summary of a walking model. It is a figure which shows a clustering result. It is a figure which shows the outline | summary of integration and division | segmentation of a laser point. It is a figure which shows the outline | summary of a traffic locus image. It is a figure which shows the walking direction calculation summary. It is a figure which shows a traffic locus image. It is a figure which shows the relationship between the distance of a left foot and a right foot.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Passer analysis device 2 ... Laser sensor 3 ... Client PC
4 ... Network 11 ... Communication processing unit 12 ... Sensing data storage unit 13 ... Parameter storage unit 14 ... Trajectory calculation unit 15 ... Trajectory data storage unit 16 ... Trajectory output unit 17 ... Display section

Claims (6)

Distance data receiving means for receiving distance data indicating a distance from a laser sensor of a passer's foot;
Sensing image data generating means for generating sensing image data indicating a detection position of the laser sensor on the spatial plane at the same time using a plurality of the distance data acquired by a plurality of the laser sensors,
Clustering means for clustering a plurality of the detection positions in the sensing image data;
A passer-by identification means for specifying a cluster corresponding to the leg of one passer-by from the clustering result;
Parameter determining means for determining a parameter consisting of a position of a foot point of the passerby, a walking direction, a stride, a walking cycle, and a walking phase from the identified passerby cluster;
A trajectory analysis processing means for analyzing a trajectory of a passerby using the parameter and the particle filter algorithm;
A passer-behavior analysis device comprising:   The passer-by-behavior analysis apparatus according to claim 1, wherein a mean-shift clustering algorithm is used in the clustering. The passerby behavior analysis apparatus according to claim 1 or 2, wherein a cluster that is obtained by dividing or integrating the cluster of the clustering results is identified as a cluster corresponding to the foot of the one passerby. The trajectory analysis processing means updates the walking phase of the parameters and updates the walking phase from the acceleration of the left foot and the right foot of the passerby identified in the clustering result cluster in the processing of the Particle filter algorithm. The travel distance of the passer-by is updated by the above, and the position of the foot point of the passer-by among the parameters is predicted from the updated walking phase, the travel distance, and the walking direction. The passer-by behavior analysis device according to claim 3. A passer behavior analysis method in a passer behavior analysis device,
Distance data reception processing for receiving distance data indicating the distance of the passer's foot from the laser sensor;
Sensing image data generation processing for generating sensing image data indicating the detection position of the laser sensor on the spatial plane at the same time using a plurality of the distance data acquired by a plurality of the laser sensors,
A clustering process for clustering a plurality of the detection positions in the sensing image data;
A passer-by identification process that identifies a cluster corresponding to the leg of one passer-by from the clustering result;
A parameter determination process for determining a parameter consisting of the position of the foot point of the passerby, the walking direction, the stride, the walking cycle, and the walking phase from the identified passerby cluster;
Trajectory analysis processing for analyzing the trajectory of passers-by using the parameters and the Particle filter algorithm,
A passer-behavior analysis method characterized by comprising: In the computer of the passer-by behavior analysis device,
Distance data receiving means for receiving distance data indicating the distance from the laser sensor of the passer's foot;
Sensing image data generating means for generating sensing image data indicating a detection position of the laser sensor on the space plane at the same time using a plurality of the distance data acquired by the plurality of laser sensors
Clustering means for clustering a plurality of the detection positions in the sensing image data;
A passerby specifying means for specifying a cluster corresponding to the leg of one passerby from the clustering result,
Parameter determining means for determining a parameter consisting of the position of the foot point of the passerby, the walking direction, the stride, the walking cycle, and the walking phase from the identified passerby cluster,
Trajectory analysis processing means for analyzing the trajectory of a passerby using the parameter and the Particle filter algorithm,
A program characterized by having executed.

Images