JP5688513B2 - Pedestrian state classification device and program - Google Patents

Description

  The present invention relates to a pedestrian state classification apparatus and program, and in particular, in a facility used by an unspecified majority such as a shopping mall, estimates the state of each pedestrian by measuring the behavior of pedestrians moving within the facility. The present invention relates to a novel pedestrian state classification apparatus and program for classification.

  The behavior of a pedestrian moving in a commercial facility shows the intention and purpose of the pedestrian. For this reason, it is considered that by analyzing the behavior of a pedestrian, the intention and the purpose of action can be estimated, and a service tailored to the purpose of the pedestrian can be provided. This is useful for both, for example, a pedestrian can obtain more detailed information about his favorite store and a facility can attract customers.

  Various analyzes have been conducted on pedestrians in facilities. In Non-Patent Documents 1 and 2, the relationship between urban spaces and pedestrians and human behavior in museums and expositions were analyzed based on the walking speed of pedestrians. In Non-Patent Document 3, pedestrian behavioral purposes in commercial facilities were classified by subject experiments using speed sensors. Non-Patent Documents 4-6 propose a method for analyzing and discriminating human behavior by paying attention to walking trajectories, and conducting research for marketing and security. In Non-Patent Document 7, pedestrians are classified according to walking speed, walking trajectory, and head direction in a museum.

Further, in Non-Patent Document 8, the present inventors conducted subject experiments on pedestrians in commercial facilities, and the interests and intentions of pedestrians looking for restaurants are shared to some extent with behavior (gaze direction and walking trajectory). It was shown that it appeared as a behavior.
Shinya Kiyota, Naoji Matsumoto, Junji Watanabe “Relationship between street space characteristics and walking speed” Architectural Institute of Japan pp949-950, 2007 Nami Kashimura, Natsuko Nagasawa, Ken Kimura, Kazuto Hayashida, Hitoshi Watanabe "Study on comparison of walking speed in viewing space" Summaries of Annual Meetings of the Architectural Institute of Japan, pp1053-1054, 2000 Tomoya Goshiro, Kazuto Hayashida, Hitoshi Watanabe "Study on the relationship between walking speed and behavioral purpose in commercial facilities" Academic Lecture Summary of Architectural Institute of Japan pp567-568, 2008 Daisuke Hanyu, Hitoshi Watanabe “Study on Inductivity of Curved Wall: Analysis from Curved Wall Shape Pattern and Walking Trajectory” Architectural Institute of Japan Annual Report pp995-996, 2003 Ichiro Toyoshima, Toyokazu Itakura, Kanako Hattori, Atsushi Yoshida, Takashi Komine “Determination of customer behavior from walking trajectory data using multi-step pattern recognition” Information Processing Society of Japan pp173-178, 200 Ichiro Toyoshima, Toyokazu Itakura, Kanako Hattori, Atsushi Yoshida, Takashi Ogura “Development of Pedestrian Behavior Discrimination Method Using Walking Trajectory” IEICE Technical Report pp19-24, 2006 Yumi Yamada, Yutaka Yanagisawa, Keiji Hirata, Tetsuji Sato “Pedestrian Action Semantics Based on Movement Trajectory and Head Direction” IPSJ Journal pp2250-2259, 2005 Kotaro Okamoto, Akira Utsumi, Daijo Yamazoe, Takahiro Miyashita, Kazuhiko Takahashi, Norihiro Hamada “Analysis of Visitor Behavior in Commercial Facilities Using Eye-Gaze Measurement” IEICE Technical Report MVE, Basics of Multimedia and Virtual Environment pp1- 6, 2009

  The above non-patent document 8 showed that the purpose of the pedestrian can be estimated through the behavior of the pedestrian, but all the pedestrians visiting the commercial facility do not have the same behavior, Behavior changes depending on purpose and degree of interest. Therefore, it is necessary to classify pedestrian behavior according to the pedestrian's internal state.

  Therefore, a main object of the present invention is to provide a novel pedestrian state classification apparatus and program.

  Another object of the present invention is to provide a pedestrian state classification apparatus and program capable of accurately classifying pedestrian states.

  The present invention employs the following configuration in order to solve the above problems. The reference numerals in parentheses, supplementary explanations, and the like indicate correspondence relationships with embodiments described later to help understanding of the present invention, and do not limit the present invention in any way.

1st invention is the non-contact sensor for detecting the position of a pedestrian, the walking speed calculation means which calculates the walking speed of a pedestrian based on the position for every predetermined time of the pedestrian detected by the non-contact sensor, non-contact Direction variation calculating means for calculating variation in the walking direction of the pedestrian based on the position of the pedestrian detected by the sensor every time, whether the walking speed is larger than the predetermined threshold and the variation in the walking direction is smaller than the predetermined threshold first condition determination means for determining whether to satisfy the first condition that, when the first condition determination means determines that satisfies the first condition, direct state classifying means for classifying the pedestrian state of the pedestrian as a direct state If the first condition determining means determines that the first condition is not satisfied, whether or not the second condition of whether or not the attention time for the specific category is greater than the threshold value is satisfied Second condition determining means for disconnection, and when the second condition determination means determines not to satisfy the second condition, and a migration status classification means for classifying the pedestrian state of the pedestrian as a migration state, walking pedestrian It is a person state classification device.

In the first invention, the pedestrian state classification device (10: reference number indicating a corresponding part in the embodiment; the same applies hereinafter) is a non-contact sensor (221-) for detecting the position of a pedestrian such as an LRF. 22n). In the embodiment, the walking speed calculation means (12,445, S3), both configured by a computer, calculates the walking speed of the pedestrian based on the position of the pedestrian detected every time by the non-contact sensor, and calculates the direction variation. The means (12,445, S3) calculates the variation in the walking direction of the pedestrian based on the position of the pedestrian detected by the non-contact sensor every predetermined time. The first condition determining means (12,445, S11) determines whether or not the first condition that the walking speed is larger than a predetermined threshold and the variation in the walking direction is smaller than the predetermined threshold is satisfied. When the first condition determining means determines that the first condition is satisfied, the straight state classification means (12,445, S13) classifies the pedestrian state of the pedestrian as a straight state. Further, when the first condition determining means determines that the first condition is not satisfied, the second condition determining means (12,445, S15) determines whether or not the second condition is satisfied, and the migratory state classification means (12,445, S17). ) Classifies the pedestrian state of the pedestrian as a migratory state when the second condition determining means determines that the second condition is not satisfied.

According to the first invention, whether the pedestrian state of the pedestrian is a direct state (the walking purpose and the destination are both determined and the location of the destination is known, and the pedestrian is heading there) or the like. Can be accurately classified. Furthermore, it is possible to accurately classify whether the pedestrian state of the pedestrian is a migratory state (for example, a pedestrian state in which the purpose of walking, the destination and the route are not determined) or the other.

The second invention is dependent on the first invention, and further comprises search state classification means for classifying the pedestrian state of the pedestrian as the search state when the second determination means determines that the second condition is satisfied. It is a state classification device.

In the second invention, the search state classification means (12,445, S21, S23) classifies the pedestrian state of the pedestrian as the search state when the second determination means determines that the second condition is satisfied. Therefore, according to the second invention, the pedestrian state of the pedestrian is accurately classified as a search state (for example, a pedestrian state where there is a purpose of walking, but a place suitable for the purpose, a route is not determined) or other than that. I can do it.

The third invention is dependent on the second invention, and the search state classification means determines whether or not the third condition of whether the pedestrian is paying attention to the specific category satisfies the third condition; When the third condition determining means determines that the third condition is satisfied, the pedestrian state classification device includes route search state classification means for classifying the pedestrian state of the pedestrian as a route search state.

In the third invention, when the third condition determination means (12,445, S19) included in the search state classification means determines that the third condition is satisfied, the route search state classification means (12,445, S21) A person state is classified as a route search state.

A fourth invention is dependent on the third invention, and the search state classification means classifies the pedestrian state of the pedestrian as a place search state when the third condition determination means determines that the third condition is not satisfied. A pedestrian state classification device including search state classification means.

In the fourth invention, when the place search state classification means (12,445, S23) included in the search state classification means determines that the third condition is not satisfied by the third condition determination means, the pedestrian state of the pedestrian is determined as the place. Classify as search state.

5th invention is a pedestrian state classification program performed by the computer of the pedestrian state classification device of a pedestrian provided with the non-contact sensor for detecting the position of a pedestrian, and a program is a non-contact computer. Walking speed calculation means for calculating the walking speed of the pedestrian based on the position of the pedestrian detected by the sensor, and the walking direction of the pedestrian based on the position of the pedestrian detected by the non-contact sensor Direction variation calculating means for calculating the variation of the first condition determining means for determining whether or not the first condition that the walking speed is larger than a predetermined threshold and the variation in the walking direction is smaller than the predetermined threshold is satisfied , the first condition when the determination unit determines that satisfies the first condition, direct state classifying means for classifying the pedestrian state of the pedestrian as orthogonal state, the first condition Second condition determining means for determining whether or not the second condition of whether or not the attention time for a specific category is greater than the threshold when the disconnecting means determines that the first condition is not satisfied, and second condition determining means Is a pedestrian state classification program that causes the pedestrian state of the pedestrian to be classified as a migration state classification means when it is determined that the second condition is not satisfied .

According to the fifth aspect , the same effect as the first aspect can be expected.

  According to this invention, since the state of the pedestrian is classified based on the measurement result such as the moving speed, the pedestrian state can be accurately classified.

  The above object, other objects, features and advantages of the present invention will become more apparent from the following detailed description of embodiments with reference to the drawings.

FIG. 1 is a block diagram showing a pedestrian state classification apparatus according to one embodiment of the present invention. FIG. 2 is an illustrative view showing a measurement range of a laser range finder (LRF) which is an example of a non-contact sensor used in the embodiment of FIG. FIG. 3 is an illustrative view showing one example of an implementation place where the pedestrian state classification apparatus of the example was tested. FIG. 4 is an illustrative view showing a pedestrian's line-of-sight direction. FIG. 5 is an illustrative view showing one example of a memory map of a RAM of the pedestrian state classification device of the embodiment. FIG. 6 is a flowchart showing an example of the pedestrian state classification operation in the pedestrian state classification device of the embodiment. FIG. 7 is a graph showing the relationship between the pedestrian state and walking speed when manually classified by experiment. FIG. 8 is a graph showing the relationship between the pedestrian state and the movement direction variation when manually classified by experiment. FIG. 9 is an illustrative view showing an example of a walking trajectory of a pedestrian. FIG. 10 is a graph showing the relationship between the pedestrian state and the stop rate when manually classified by experiment.

  Referring to FIG. 1, a pedestrian state classification apparatus 10 according to an embodiment of the present invention includes a computer 12, and the computer 12 includes other internal memory 14 such as RAM and ROM, and another network 18 as necessary. A communication device 16 is provided for communicating with a computer (not shown) of the system or apparatus. However, the communication device 16 may not be built in the computer 12. The memory 20 connected to the computer 12 may be, for example, a hard disk or a large-capacity semiconductor memory, and may be used for storing a program for processing by the computer 12 and data before and after processing.

  The computer 12 is further connected to an LRF 221-22n and a video camera 241-24m. When it is not necessary to distinguish the LRFs 221-22n, they are collectively referred to as “LRF22”, and when it is not necessary to distinguish the video cameras 241-24m, they are collectively referred to as “cameras 24”.

  The LRF 22 measures the distance to the object or the person from the time it takes to irradiate the laser and reflect it back to the object or the person. For example, a laser irradiated from a transmitter (not shown) is reflected by a rotating mirror (not shown), and the front is scanned in a fan-like manner at a certain angle (for example, 0.5 degrees), and the reflected wave returns. By detecting the time until, the distance to the object can be known. As an example, LRF (model LMS200) manufactured by SICK can be used. When this LRF 22 is used, the position of the pedestrian can be measured with an error of about 6 cm, for example.

  As shown in FIG. 2, the measurement range of the LRF 22 is indicated by a semicircular shape (fan shape) having a radius R (R≈8 m). That is, the LRF 22 can measure the direction of 90 ° left and right within a predetermined distance (R) when the front direction is the center. And as shown, for example in FIG. 3, LRF22 is arrange | positioned in various places. For example, in FIG. 3, each of the LRFs 221 to 226 is arranged so that the detection areas overlap. The reason is that the position of each pedestrian is detected by distance data detected by two or more LRFs for the same pedestrian. A method for measuring the position of a person or an object using the LRF 22 as described above is well known, for example, in Japanese Unexamined Patent Application Publication No. 2009-168578.

  In addition, in order to enable analysis on the detailed behavior of pedestrians, in the implementation place 26 shown in FIG. 3, a plurality of the entire place 26 (5 in the embodiment) are combined with the position measurement by the LRF 221-227 described above. The video camera 241-245. By applying a gaze estimation process, which is described in detail in Japanese Patent Application Laid-Open No. 2008-102902 related to the application of the present applicant, for example, to the face images photographed by these cameras 24, it is shown in FIG. Such a sight line direction of each pedestrian, that is, a direction facing the face can be detected.

3, in this example, outlets 30 are set at both ends of the passage 28, and a corner 32 is formed in front of one outlet 30 of the passage 28. Further, a first restaurant 34, a second restaurant 36, and a third restaurant 38 are arranged on both sides of the passage 28. These exit 30, turn 32, restaurant 34-38, and guide plate 40 in FIG. 3 are called “target” in the sense of an object of interest, and are indicated by the symbol “p _j ” in the mathematical formulas described below. Each of these targets 30-40 can be classified into categories, and these categories are indicated by the symbol “Cp _j ” in the mathematical formulas described later.

  FIG. 5 shows an example of a memory map of the RAM 42 included in the internal memory 14 of the computer 12 in the embodiment of FIG. 1. In this embodiment, the RAM 42 includes a program area 44 and a data storage area 46. In the program area 44, an OS area 441 for storing the OS is formed. Further, the program area 44 is provided with a distance data acquisition program storage area 442 for storing a program (distance data acquisition program) for acquiring distance data of each LRF for each pedestrian. Furthermore, a position data acquisition program storage area 443 for storing a program (position data acquisition program) for acquiring the position data of each pedestrian from the distance data of each LRF for each pedestrian is provided. In addition, a distance gaze detection program storage area 443 for storing a program (gaze detection program) for detecting the gaze direction of each pedestrian from each pedestrian's face image and each pedestrian's face image based on the detected gaze. A face direction data acquisition program storage area 444 for storing a program (face direction data acquisition program) for acquiring data of the direction in which the camera is facing is provided. The classification program set in the classification program storage area 445 is a program for classifying the pedestrian state of each pedestrian based on each position data and face direction data detected by these programs. In this classification program 445, calculation formulas shown in the following equations 1 to 9 and the like necessary for pedestrian state classification are also set.

  The data storage area 46 includes, for example, a sensor data storage area 461 for temporarily storing distance data of the LRF 22 and video data of the video camera 24, a map data storage area 462 for storing map data, and a histogram storage area. 463 and a working area 464 are set. For example, as shown in FIG. 3, the map data is arranged in the location data indicating the location of each target in the implementation location, the category of each target (type of each target, restaurant, coffee shop, toilet, store, etc.), and the implementation location. In the embodiment, position data of the installation position of the LRF 22 and the video camera 24 is stored. In this embodiment, the histogram stored in the histogram storage area 463 is assumed to be a histogram of a target (attention target) focused by each pedestrian or a histogram (table) of a target category (attention category) focused on. . If you look at the histogram of the target of interest, you can see the number of times of attention (attention time) for each target of each pedestrian, and if you look at the histogram of the category of attention, you can see the number of times of attention (attention time) for each category of target of each pedestrian. I understand.

Map data mentioned above is basically set in the map data storage area 462 is issued to read from the external memory 20 (FIG. 1) depending on the implementation location. Similarly, the program may be set in the program area 44 from the external memory 20 whenever necessary.

  Before the detailed description of the operation of the embodiment performed with reference to FIG. 6, the background of the present invention will be briefly described.

  There is a possibility that information directly related to the consumption of goods or services can be obtained from the behavior of pedestrians visiting a commercial facility, and the facility side or each store side is highly interested. For this reason, various methods for analyzing customer behavior have been proposed and used in practice. Among them, one of the analysis methods that is currently widely used is POS (point-of-sale information management). The information obtained from POS is not limited to the sales volume of each product, but is linked to the consumption linkage between products, the sex of customers, the relationship between age groups and consumption, etc. Playing a role. However, the information obtained by POS is only information related to “consumed” products and services, and does not give information on “pre-consumption” behavior that the customer was interested in but did not consume. For this reason, “pre-consumption behavior” has been analyzed to strengthen and supplement POS.

  Therefore, in the embodiment described here, as an example of analysis of “pre-consumption behavior”, a person who is visiting a store on the floor of a commercial facility is tracked using an LRF or video shooting of the floor of the commercial facility as an example. The range of human interest and intention was estimated based on the behavior during walking such as speed and walking trajectory, and the following pedestrian behavior and pedestrian state were assumed.

The behavior of pedestrian "or walking purpose is definite", classified by "whether the location corresponding to the target (destination) is definite", "whether or not there is knowledge of the directions to the destination", no walking purpose The pedestrian state of a pedestrian is called a “walking state”, and the pedestrian state of a pedestrian looking for a place suitable for the purpose or a pedestrian looking for a fixed destination is “search state (location or direction)” The pedestrian state of a pedestrian who knows all the purpose of walking, the destination, and the route is referred to as “direct state”. If the pedestrian state of a pedestrian can be estimated and classified, the following services suitable for pedestrians in each state can be provided.

(1) Migrating state This is a pedestrian state where the purpose of walking, destination, and directions are not fixed, and stays on the floor (such as the place 26 shown in Fig. 3) for waiting time and meeting people. The pedestrian who falls is in this state. Services such as recommended spots on the floor and information on events are assumed.

(2) Searching status Pedestrians who have a purpose of walking but are in a pedestrian state where the purpose and directions are not determined and are searching for the floor looking for a restaurant or target store Corresponds to this state. A service that provides information on a store (category) corresponding to a walking purpose is assumed. Also, pedestrians who have both a walking purpose and a destination but who are searching for a destination without knowing the location of the destination are also classified into this state. Services such as providing route information from the current location to the destination are assumed.

(3) Direct state The purpose of walking and the destination are both determined, the location of the destination is known, and the pedestrian is heading towards it. For pedestrians who are in a “straight state”, it may be possible to provide detailed information such as product introduction after conducting detailed analysis of the purpose of visiting the store by measuring the face direction and gaze near the destination.

  Based on the above background, the operation of the embodiment of FIG. 1 will be described below together with the flowchart of FIG. The pedestrian state classification operation shown in the flowchart of FIG. 6 is executed for all pedestrians existing in the place 26 (FIG. 3) once in a short time such as 0.5 seconds as an example. That is, the pedestrian states of all the pedestrians at the implementation place 26 are classified every predetermined time.

In the first step S1 of FIG. 6, the computer 12 detects the position (xtn) of each pedestrian at time (tn) according to the distance data acquisition program 442 and the position data acquisition program 443 shown in FIG. The face direction (wtn) of each pedestrian at time (tn) according to the program 444 and the face direction data acquisition program 445 is detected. That is, since the installation position of each LRF 22 can be known from the map data, the position (xtn) of the pedestrian can be specified by converting the distance between the pedestrian detected by each LRF into the position data. Similarly, since the installation position of the video camera 24 and the direction in which the video camera 24 is facing can be understood from the map data, the pedestrian's face direction (wtn) detected by the video camera can be represented in the global (overall) coordinate system of the place of implementation 26. Thus, as described later, it can be determined whether the face of the pedestrian is facing the target, that is, whether the pedestrian is paying attention to the target.

In the subsequent step S3, the computer 12 follows the classification program 445 (FIG. 5) to determine the walking (movement) speed (vt _n ) and average speed (vt _n ) of each pedestrian at time (t _n ). Bar) and velocity vector variation (dt _n ) are calculated based on Equation 1, Equation 2, and Equation 3, respectively. “Bar” is a horizontal bar (−) indicating an average value to be written on a character (in this case, “v”).

  In Equation 1, since the difference between the previous position and the current position is divided by the difference between the previous time and the current time, the movement speed is obtained. In the first half of Equation 2, the movement speed obtained in this way is the nearest N + 1 times. Since the sum of the values is divided by N + 1, an average velocity vector is obtained. In the second half of Equation 2, the covariance matrix for the velocity vector is calculated.

Note that the first half (vt _n hat) of Equation 3 is an average value for the latest N + 1 times of the magnitude of the velocity vector, and indicates the average velocity. The latter half of Equation 3 is calculated as a trace of the covariance matrix (the sum of the diagonal components of the covariance matrix) obtained by calculating the magnitude of the velocity vector variation in Equation 2. “Hat” is a mountain shape (^) indicating an average value to be written on a letter (in this case, “v”).

In this manner, in step S3, the walking speed (vt _n ), average speed (vt _n hat), and speed vector variation (dt _n ) of each pedestrian at that time are calculated.

Subsequently, in step S5, the computer 12 determines whether or not the distance between each pedestrian and the target (such as a restaurant, an exit, or a corner if the place of implementation in FIG. 3) is short. That is, it is determined whether the inequality of Equation 4 is satisfied. That is, in step S5, it is determined whether or not the distance between the pedestrian position (vt _n ) and the target position (xp _j ) is smaller than the threshold (εp). Since the target position is set in the map data described above, it may be used.

If “YES” is determined in the step S5, the computer 12 determines whether or not the pedestrian is paying attention to the target based on the equation 5 in the next step S7. In Equation 5, the inner product of the face direction (wt _n ) at that time and the target direction calculated in step S5 is calculated, and whether or not it is larger than the threshold value cos θ, that is, the pedestrian's face direction faces the target Judging whether it is close to.

If “YES” in the step S5 and “YES” in the step S7, that is, if it is determined that the pedestrian is approaching any target and paying attention, the computer 12 in the next step S9, the histogram storage area The histogram of the target of attention (p _j ) and the category of attention (Cp _j ) of 463 (FIG. 5) is updated (Equation 6).

  Thereafter, the process proceeds to step S11. However, if “NO” is determined in step S5 or if “NO” is determined in step S7, the process skips step S9 and proceeds to step S11.

  In step S11, the computer 12 determines whether or not the walking (moving) speed of the pedestrian is larger than a predetermined threshold, for example, 0.5 m / s and the variation in the speed vector is smaller than the predetermined threshold, based on Equation 7. It is determined whether or not the first condition is satisfied. The fact that the walking speed is high and the variation in the speed vector is small, it can be determined that the pedestrian state of the pedestrian is the “direct state” described above. That is, it is a state where it is walking fast toward the target. Therefore, if “YES” is determined in step S11, the computer 12 classifies the pedestrian state of the pedestrian as “directly-directed state” in step S13. Therefore, step S13 constitutes an orthogonal state classification means.

When “NO” is determined in step S11, the walking speed is not large or the variation in the speed vector is not small. That is, the first condition is not satisfied. In this case, it can be assumed that the pedestrian is walking slowly or the walking direction is not settled (disjoint). Therefore, in the next step S15, the computer 12 determines whether or not the attention time to the specific category of the pedestrian who is not in the direct state is larger than the threshold, that is, whether the second condition is satisfied, according to Equation 8. In Expression 8, the average attention time obtained by dividing the attention time (B | Cp _j |) of the category indicated by the histogram of the attention category by the number of times (n) is larger than a predetermined threshold, that is, which pedestrian is It is determined whether or not the category has been watched for more than the threshold time. “NO” in step S15 means that the second condition is not satisfied, and that the pedestrian has a short attention time for any category and the attention target is not consistent. In such a pedestrian state, in the embodiment, it is assumed that the state is the “walking state” described above. Therefore, when “NO” is determined in the step S15, the computer 12 classifies the pedestrian state of the pedestrian as the “walking state” in a step S17. Therefore, step S17 constitutes a migration state classification means.

  When it is neither a direct state nor a migratory state, that is, when “NO” is determined in step S11 and “NO” is determined in step S15, that is, when the first condition is not satisfied and the second condition is satisfied, In the next step S19, the computer 12 determines whether the pedestrian is paying attention to a specific category, that is, whether the third condition is satisfied. In this embodiment, since determination is made based on Equation 9 in step S19, in step S19, it is determined whether or not the pedestrian is paying attention to the exit 30 or the corner 32 (FIG. 3) for a long time.

  If “YES” in the step S19, that is, if the third condition is satisfied, the pedestrian is paying attention to the exit 30 and the turning corner 32 for a long time. In the next step S21, the computer 12 walks at that time. A person's pedestrian state is classified as a state of walking while searching for directions (direction search state). Therefore, step S21 constitutes a route search state classification means.

If “NO” in the step S19, that is, if the third condition is not satisfied, the pedestrian is in a state of paying attention to a target in a category other than the exit 30 and the corner 32 for a long time. Therefore, in this case, since the attention category is, for example, a store that is not related to the route search, in the next step S23, the computer 12 determines that the pedestrian state of the pedestrian is the target of the specific category (Cp _j ) It is classified as a state of walking while searching for a place (location search state). Therefore, step S23 constitutes a place search state classification means.

  However, when the state is neither the direct state nor the migratory state, it is a target search state in which a certain target is being searched. Therefore, from this point of view, steps S21 and S23 can also be regarded as search state classification means.

  As described above, according to the described embodiment, the pedestrian state of each pedestrian can be obtained by a non-contact sensor such as the LRF 22 or the video camera 24 and without attaching any attachment such as a wireless tag to the pedestrian. Can be classified. Here, experiments conducted by the inventors are shown, and the validity of such classification is explained.

  In the experiment, we first extracted the daytime data in which there are pedestrians in the above-mentioned state, direct state, migratory state, and search state from the data of about 5000 people on weekdays, and performed manual labeling according to the following criteria for each classification. Sample data was obtained. First, 10 pedestrians who have been in the measurement floor for more than a certain time (180 seconds or more) were classified as “walking state”. Next, 26 pedestrians who entered the restaurant in the measurement area without stopping on the way were regarded as “directly running”. Of the other data, pedestrians who passed (passed through) the measurement area were excluded, and 21 pedestrians who stopped on the way or stopped by a guide board were classified as “search state”.

And the distribution of the average walking speed (vt _n hat) about the pedestrian of each state is shown in FIG. As can be seen here, the speed of the pedestrian in the direct state is greater than in the other two states. A pedestrian in a straight state has a clearer purpose of walking than a pedestrian in search and excursion, and this is considered to be reflected in a difference in walking speed. Statistically significant differences were observed between the pedestrians in the direct state and the search state (p <0:01), and between the pedestrians in the direct state and the migratory state (p <0:01).

Subsequently, the walking trajectory, in particular, the movement direction variation shown in FIG. 8 was compared. The variation (dt _n ) in the moving direction was defined as in Equation 3 based on the velocity vector (vt _n ) at each time. An example of the walking trajectory is shown in FIG. FIG. 9 also shows a characteristic walking locus depending on each pedestrian state.

  FIG. 8 shows variations in the moving direction for each state. From FIG. 8, pedestrians in the straight state have small variations in the moving direction and the moving direction is relatively stable, whereas pedestrians in the search state and the migratory state have large variations in the moving direction and frequently move in the moving direction. You can see that Statistically significant differences were observed between the pedestrians in the direct state and the search state (p <0:01), and between the pedestrians in the direct state and the migratory state (p <0:01).

With respect to the extracted three-state pedestrians, an analysis was made as to the proportion of the stop state in which the walking speed is below a certain level within the measurement time. The stopping rate (rt _n ) is expressed by Equation 10. Here, (v ₀ ) is a threshold value (for example, 0.5 m / s) as to whether or not to consider a stop, and (rt _n ) indicates a ratio of the number of times considered to be a stop in the past N + 1 observations.

  FIG. 10 shows the stop rate of each pedestrian in three states. It can be seen from FIG. 10 that the pedestrian in the straight state hardly sees the stop state. On the other hand, it can be seen that the pedestrian in the migratory state takes the action of stopping for a long time in addition to the low walking speed when moving as described above. Between pedestrians in direct state and search state (p <0:01), Between pedestrians in direct state and migratory state (p <0:01), Between pedestrians in search state and migratory state (p <0:01) There were statistically significant differences in each.

Furthermore, the gait or attention behavior of the pedestrian is deeply related to the interest and intention of the pedestrian. When a pedestrian takes action such as shaking his head, the interest of the pedestrian has not been determined and the interest has been transferred. In this analysis, pedestrians in each state were compared by counting the number of times the face was rotated large (about 45 degrees) relative to the direction of travel by visual analysis using video. Table 1 shows the average number of swings per minute for pedestrians in each state. From the results in Table 1, there was a large difference between the migratory state and the others, but no significant difference was found in the search state or the direct state.

  Furthermore, the inventors used the walking speed, the variation of the walking trajectory, and the stoppage rate in the floor using the statistically significant difference observed on weekday (Thursday) daytime (12: 00-14: 00), evening ( 18: 00 to 19: 00), and the data of three time zones of daytime (12: 00 to 14:00) on holidays (Sunday). For the classification, the nearest neighbor method (NN method) was used, and classification was made into three states: an orthogonal state, a search state, and a migratory state.

  As a result, among 2102 pedestrians during the daytime hours on weekdays, 1640 (78%) were classified as direct, 349 (17%) were searched, and 113 (5%) were classified as migrating. . During weekday evening hours, the percentage remains almost unchanged, out of 747 pedestrians, 615 (82%) are in a direct state, 101 (14%) are in a search state, and 31 (4%) are in a migratory state. It was classified into.

  On the other hand, in the daytime hours on holidays, out of 4671 pedestrians, 2747 (59%) are classified as direct, 1778 (38%) are searched, and 146 (3%) are classified as migrating. As a result, the ratio of search states increased. This result can be interpreted as follows.

  The shopping mall that was tested this time is located between the nearest station on the railway (new tram) and the office building, and is used as a route for office workers to go to work and return home. Also, during the day, these office building workers walk a lot in this area to have lunch or head for their destinations. Since these pedestrians are familiar with the locations of the shopping mall passages and stores, it is considered that the number of pedestrians in the “direct state” increased on weekdays regardless of the time zone. On the other hand, we observe from video images that there are many pedestrians who are unfamiliar with this facility, such as families who visited shopping malls for the purpose of eating and shopping on weekends and event participants who participate in events held around shopping malls. It was done. These pedestrians may not be able to accurately grasp the location of the destination or the destination has not been determined. For this reason, compared with a weekday, the ratio of the pedestrian of a direct state decreases on a holiday, and the possibility that the pedestrian of a search state will increase is high. The classification results obtained above are consistent with these observation results, and it can be said that the results suggesting the validity of the classification of the examples described above were obtained.

  In addition, if the pedestrian state classified as shown in FIG. 6 is used, an event, a service, or the like that exactly matches each pedestrian state can be provided. For example, guidance information for each pedestrian and recommended information for attracting pedestrians to each target can be distributed via the network 18 using the communication device 16 shown in FIG.

  In the above-described embodiment, the LRF is used to detect the position of a pedestrian, but an ultrasonic sensor, an infrared sensor, or the like can also be used.

10 ... Pedestrian state classification device 12 ... Computer 14 ... Internal memory 22, 221-22n ... LRF
24, 241-24m ... Video camera

Claims (5)

A non-contact sensor for detecting the position of a pedestrian,
Walking speed calculation means for calculating the walking speed of the pedestrian based on the position of the pedestrian detected every predetermined time detected by the non-contact sensor;
Direction variation calculating means for calculating variation in the walking direction of the pedestrian based on the position of the pedestrian detected every predetermined time detected by the non-contact sensor,
First condition judging means for judging whether or not a first condition is satisfied that the walking speed is larger than a predetermined threshold and the variation in the walking direction is smaller than the predetermined threshold ;
Straight state classification means for classifying the pedestrian state of the pedestrian as a straight state when the first condition determination means determines that the first condition is satisfied ;
Second condition determining means for determining whether or not a second condition of whether or not an attention time for a specific category is greater than a threshold when the first condition determining means determines that the first condition is not satisfied; and
A pedestrian state classification device for pedestrians, comprising: a traveling state classification unit that classifies a pedestrian state of the pedestrian as a traveling state when the second condition determining unit determines that the second condition is not satisfied . When said second determination unit determines that satisfies the second condition, the further comprising a search condition classifying means for classifying the pedestrian condition of the walker as the search condition, the pedestrian status classification device according to claim 1. The search state classification unit includes a third condition determination unit that determines whether or not a third condition of whether or not a pedestrian is paying attention to a specific category, and the third condition determination unit determines whether the third condition is satisfied. The pedestrian state classification device according to claim 2 , further comprising: a route search state classification unit that classifies the pedestrian state of the pedestrian as a route search state when it is determined that the pedestrian is satisfied. The search state classification unit includes a location search state classification unit that classifies a pedestrian state of the pedestrian as a location search state when the third condition determination unit determines that the third condition is not satisfied. 3. The pedestrian state classification device according to 3. A pedestrian state classification program executed by a computer of a pedestrian state classification device for a pedestrian comprising a non-contact sensor for detecting the position of the pedestrian, the program comprising:
Walking speed calculation means for calculating the walking speed of the pedestrian based on the position of the pedestrian detected every predetermined time detected by the non-contact sensor;
Direction variation calculating means for calculating variation in the walking direction of the pedestrian based on the position of the pedestrian detected every predetermined time detected by the non-contact sensor,
First condition judging means for judging whether or not a first condition is satisfied that the walking speed is larger than a predetermined threshold and the variation in the walking direction is smaller than the predetermined threshold ;
Straight state classification means for classifying the pedestrian state of the pedestrian as a straight state when the first condition determination means determines that the first condition is satisfied ;
Second condition determining means for determining whether or not a second condition of whether or not an attention time for a specific category is greater than a threshold when the first condition determining means determines that the first condition is not satisfied; and
A pedestrian state classification program that functions as a traveling state classification unit that classifies a pedestrian state of a pedestrian as a traveling state when the second condition determining unit determines that the second condition is not satisfied .

Images