JP5563796B2 - Pedestrian movement estimation apparatus and method - Google Patents

Description

  The present invention relates to a pedestrian movement estimation apparatus and method for estimating movement behavior of a pedestrian.

  As a conventional pedestrian movement estimation apparatus and method, for example, those described in Patent Document 1 are known. In the thing of patent document 1, a pedestrian is detected from the pixel pattern information etc. of a moving object, both leg support time and the maximum stride of the pedestrian are obtained, and whether or not the pedestrian stops based on these information The future position of the pedestrian is predicted using the determination result, and the range where the pedestrian can exist is specified.

JP 2009-12521 A

  However, the following problems exist in the prior art. That is, in an urban area with many pedestrians, pedestrians often move more rapidly than vehicles. For this reason, the estimation accuracy cannot be increased by the method of linearly estimating the movement behavior of the pedestrian from the movement immediately before the pedestrian.

  The objective of this invention is providing the pedestrian movement estimation apparatus and method which can estimate the movement action of a pedestrian with high precision.

The pedestrian movement estimation apparatus of the present invention includes movement information acquisition means for acquiring movement information of a pedestrian and movement information of at least one moving body existing around the pedestrian, movement information of the pedestrian, and movement of the moving body. Based on the information, distance-related information acquisition means for acquiring information about the distance between the pedestrian and the moving body, speed-related information acquisition means for acquiring information about the walking speed of the pedestrian, and information about the destination of the pedestrian A destination-related information acquisition means for acquiring, an estimation means for estimating the movement behavior of the pedestrian using information on the distance between the pedestrian and the moving body and information on the destination of the pedestrian , and the estimation means comprises: Estimate pedestrian movement behavior using a model that defines pedestrian behavior based on information on distance between pedestrian and moving body, information on pedestrian destination and information on pedestrian walking speed Characterized in that that.

In general, a pedestrian has a habit of keeping a distance from a moving body such as another pedestrian at a certain level or more. Therefore, the movement information of the pedestrian and the movement information of at least one moving body existing in the vicinity of the pedestrian are acquired, information on the distance between the pedestrian and the moving body is acquired based on the movement information, and the walking By estimating the pedestrian's movement behavior using a model that defines the pedestrian's behavior based on information about the distance between the pedestrian and the moving body, information about the pedestrian's destination, and information about the pedestrian's walking speed , For example, even when the movement of a pedestrian changes rapidly in response to a moving body such as another pedestrian, the movement behavior of the pedestrian can be appropriately estimated. Thereby, the estimation precision of a pedestrian's movement action can be improved. Moreover, the pedestrian usually tries to maintain the traveling direction to the destination as it is. Therefore, when estimating the movement behavior of the pedestrian, the movement behavior of the pedestrian can be estimated with higher accuracy by considering information on the destination of the pedestrian.

The pedestrian movement estimation method of the present invention is a pedestrian movement estimation method by the pedestrian movement estimation device, and the movement information acquisition means of the pedestrian movement estimation device exists in the vicinity of the pedestrian movement information and the pedestrian. The step of acquiring movement information of at least one moving body, and the distance-related information acquisition means of the pedestrian movement estimating device is based on the movement information of the pedestrian and the movement information of the moving body, and the distance between the pedestrian and the moving body. A step of acquiring information about, a step of acquiring speed related information of the pedestrian movement estimation device, a step of acquiring information of the walking speed of the pedestrian, and a destination related information acquisition unit of the pedestrian movement estimation device using acquiring information related to destinations, means for estimating the pedestrian motion estimation device, and information about the destination of the pedestrian information about the distance between the moving entity and pedestrians, pedestrian Estimating the movement behavior, and in the step of estimating the movement behavior of the pedestrian, the estimation means includes information on the distance between the pedestrian and the moving body, information on the destination of the pedestrian, and the walking speed of the pedestrian. The movement behavior of the pedestrian is estimated using a model that defines the behavior of the pedestrian based on the information regarding the pedestrian .

  By implementing such a pedestrian movement estimation method, as described above, even when the movement of a pedestrian changes rapidly in response to a moving body such as another pedestrian, the movement behavior of the pedestrian is appropriately Can be estimated. Thereby, the estimation precision of a pedestrian's movement action can be improved.

  According to the present invention, it is possible to estimate the movement behavior of a pedestrian with high accuracy. Accordingly, for example, when tracking a pedestrian using a technique for estimating a pedestrian's moving behavior, the tracking performance can be improved.

It is a block diagram which shows schematic structure of one Embodiment of the pedestrian movement estimation apparatus concerning this invention. It is a graph which shows an example of the relationship between the distance of pedestrians and discomfort. It is a graph which shows the other example of the relationship between the distance of pedestrians, and discomfort. It is the schematic which shows an example of the image obtained by image | photographing the lower building entrance vicinity from the window of the upper floor of a building. It is a graph which shows an example of the discomfort degree number when changing the speed vector of the next step of a pedestrian. It is a figure which shows an example of the movement path | route of a pedestrian detected by image recognition. It is a flowchart which shows the procedure of the pedestrian movement estimation process by the pedestrian movement estimation part shown in FIG.

  Preferred embodiments of a pedestrian movement estimation apparatus and method according to the present invention will be described below in detail with reference to the drawings.

  FIG. 1 is a block diagram showing a schematic configuration of an embodiment of a pedestrian movement estimation apparatus according to the present invention. In the figure, a pedestrian movement estimation apparatus 1 according to the present embodiment is an apparatus that detects a pedestrian by image recognition, predicts a movement path of the pedestrian, and tracks the pedestrian.

  The pedestrian movement estimation device 1 includes a camera 2 that images a pedestrian and an ECU (Electronic Control Unit) 3 connected to the camera 2. The camera 2 may be fixedly installed at a high place such as a building, or may be mounted on a moving body such as a vehicle.

  The ECU 3 includes a CPU, a memory such as a ROM and a RAM, an input / output circuit, and the like. The ECU 3 includes an image processing unit 4, a storage unit 5, and a pedestrian movement estimation unit 6.

  The image processing unit 4 performs image processing such as filter processing, binarization processing, and feature extraction processing on the captured image of the pedestrian acquired by the camera 2 to generate an image frame (image data). The storage unit 5 stores parameters used for arithmetic processing by the pedestrian movement estimation unit 6.

  The pedestrian movement estimation unit 6 performs a predetermined calculation process based on the image frame generated by the image processing unit 4 to predict the path of the pedestrian, and further tracks the pedestrian. At this time, the pedestrian movement estimation unit 6 performs the course prediction and tracking of the pedestrian using a model that defines the behavior of the pedestrian.

  First, a model that defines pedestrian behavior will be described. The point of predicting the path of the pedestrian is that the behavior of the pedestrian is formulated as an energy function for obtaining the next step of the pedestrian.

The next step of the pedestrian is represented by a velocity vector v starting from the current pedestrian position. The velocity vector v includes speed and direction information. The velocity vector v in the unit time t _s seconds only advanced pedestrian position becomes the predicted position of the pedestrian after the unit time t _s. Unit time t _s is the movement of pedestrians is short enough that can be assumed to be a linear motion, for example, 0.4 seconds. It predicts the pedestrian position after the unit time t _s, by repeating the like to predict the position of the further unit time t _s s the position of the predicted pedestrian as a base point, a pedestrian to an arbitrary time later Can be predicted.

The movement of a pedestrian can be expressed as a movement on a two-dimensional plane. Therefore, the velocity vector v of the pedestrian, the speed v _x on the x-y plane _is represented by two elements of v _y. Here, the i-th pedestrian and P _i, the velocity vector v _i of the pedestrian P _i is expressed as follows. Here, v _ix and v _iy are the velocities of the pedestrian P _{i on} the x axis and the y axis.
v _i = [v _ix , v _iy ] ^T

Now, let the velocity vector of the next step of the pedestrian P _{i be} v _i ′, and define an energy function for obtaining the next step of the pedestrian P _i as follows.
E _i (v _i ′) = λ ₀ I _i (v _i ′) + λ ₁ S _i (v _i ′) + λ ₂ D _i (v _i ′) (A)

The energy function E _i is a weighted linear sum of three energy terms I _i , S _i , D _i . The weighting coefficient of each energy term is λ ₀ , λ ₁ , λ ₂ . Determining the velocity vector v _i ′ that minimizes the energy function E _i determines the next step of the pedestrian P _i .

Next, functions I _i , S _i and D _i will be described in detail. Here, the following three conditions are assumed as conditions for determining pedestrian behavior.

(1) The distance between pedestrians is kept above a certain level In general, pedestrians feel uncomfortable if they are too close to other pedestrians, so keep a certain distance between them. At last. Discomfort is highest when pedestrians are in close contact (when the distance between each pedestrian is zero), and the discomfort decreases rapidly as the distance between the pedestrians increases. When it comes to the distance, almost no discomfort is felt.

If the distance between the pedestrian P _i and another pedestrian P _j is d _ij and the discomfort (number of discomforts) is F _ij , the distance d _ij between the pedestrians P _i and P _j and the discomfort F _ij The relationship is expressed as a normal distribution function as shown in FIG. The normal distribution function at this time can be expressed as follows, for example.
F _ij = e ^ (− (d _ij ² / 2σ _d ² ))

Here, σ _d is a parameter indicating the spread of the distribution, and can be appropriately set according to the nature and location of the pedestrian. For example, if it is a crowded place such as a station in the morning and evening, the distance between pedestrians cannot be increased, and even if the distance between pedestrians is close, it must be allowed, so σ _d is set small. Should. When the number of pedestrians is small in a wide place, σ _d should be set large. Σ _d can also reflect the characteristics of race and ethnicity.

Note that the function applied to the relationship between the distance d _ij between the pedestrians P _i and P _j and the discomfort F _ij is not particularly limited to the normal distribution function as described above, and FIG. It may be a non-linear function as shown in FIG. 3 or a linear function as shown in FIG.

The above is only about the relationship between a pair of pedestrians (pedestrian P _i and pedestrian P _j ), but in reality, it is necessary to consider the relationship between all the pedestrians around the pedestrian P _i . is there. Therefore, the pedestrian P _i other than the other of the pedestrian and P _{r (r} ≠ i), the sum of the discomfort F _ir generated between the pedestrian P _i and all of the other pedestrian P _r, of the following Define as follows.

However, w _r (i) is a parameter for adjusting the discomfort of the pedestrian P _r with respect to the pedestrian P _i . W _r (i) may be set to a constant value (for example, 1) for all pedestrians, or may be set in consideration of the distance between the two persons, the angle difference in the traveling direction, or the like. When considering the distance and the angle difference in the traveling direction, for example, the definition is as follows.
w _r (i) = w _r ^d (i) * w _r ^φ (i)
^{_{w r d (i) = e}} ^ (- (d ir 2 / 2σ w 2))
w _r ^φ (i) = ((1 + cos φ) / 2) ^β

Here, d _ir is the distance between the pedestrians P _i and P _r , and φ represents the angular difference in the traveling direction of the pedestrians P _i and P _r . The above formula shows that for a set of pedestrians, the weight coefficient w _r (i) increases as the distance from each other is closer and toward each other. If the distance between the two people is already short, an increase in discomfort is expected, so it is considered that the next step will not go further, and if the two people are facing the same direction This is because, similarly, an increase in discomfort is expected, and it is considered that the next step will not go in that direction.

Thus, energy term I _i will mean the sum of the discomfort that occurs between the pedestrian P _i and the pedestrian P _r of the surrounding. Then, a model is established in which the speed vector v _i ′ for the next step of the pedestrian P _i is determined so that the energy term I _i is minimized. Since the energy term I _i is a function that varies depending on the velocity vector v _i ′, it is correctly described as I _i (v _i ′) as in the above equation (A). Further, d _ij is a distance between the pedestrians P _i and P _j after a unit time t _s seconds as the next step.

By defining the energy term I _i (v _i ′) as described above, human social behavior characteristics that minimize discomfort caused by the distance between the pedestrian and other pedestrians in the vicinity. You can build a mathematical model that fits In addition, unlike a vehicle, a pedestrian may walk while touching (such as a shoulder touching). Since the energy term I _i (v _i ′) is a model that can also allow contact between pedestrians, it is close to the actual pedestrian behavior by adjusting parameters such as σ _d according to the degree of congestion, etc. A model can be provided. Furthermore, by adjusting parameters such as σ _d according to the characteristics of the place, the present invention is applicable even when the tendency of pedestrian behavior such as a narrow road and a plaza is different.

(2) A pedestrian tries to maintain a constant speed Generally, a pedestrian has the property of trying to maintain a constant walking speed, and feels uncomfortable as the deviation from the walking speed increases. If the preferred walking speed of the pedestrian P _i is u _i , the discomfort degree number S _i (v _i ′) is defined as follows.

Here, the favorite walking speed u _i of the pedestrian P _i may be an average speed of a plurality of pedestrians measured in advance according to the environment, or may be a speed immediately before the pedestrian P _i , The average speed for the past several seconds may be used, or the average speed of surrounding pedestrians may be used.

By defining the energy term S _i (v _i ′) as described above, it is possible to construct a model that reflects human behavioral characteristics in which a pedestrian tries to maintain his / her walking pace. Moreover, the walking speed of a pedestrian is hardly influenced by the speed of surrounding objects like a vehicle. For this reason, by setting the energy term S _i (v _i ′) independent of the energy term I _i (v _i ′), a model close to the actual movement can be constructed.

(3) A pedestrian tries to go to a destination Generally, a pedestrian tries to go to a specific destination, and discomfort increases as the distance from the destination increases. When the destination of the pedestrian P _i is z _i and the current position of the pedestrian P _i is k _i , the discomfort degree number D _i (v _i ′) is defined as follows.

Here, the destination z _i may be set according to the environment, or may be set on an extension line in the direction in which the pedestrian P _i has been walking. When the destination z _i is set according to the environment, for example, if there is a building entrance nearby, the entrance is set as the destination z _i . Further, at the pedestrian crossing, the end of the pedestrian crossing is set as the destination z _i . In this case, the destination z _i can be set more correctly by using the map information.

By defining the energy term D _i (v _i ′) as described above, a model reflecting the pedestrian's basic behavioral characteristic that the pedestrian moves toward the destination can be constructed. In addition, since the destination z _i can be freely defined, it is closer to the actual pedestrian movement compared to a vehicle prediction model that has restrictions such as driving on the road or obeying traffic rules. A model can be built.

An energy function E _i that is a linear sum of weights of the energy terms I _i (v _i ′), S _i (v _i ′), and D _i (v _i ′) defined in the above items (1) to (3). (v _i ′) is a model for predicting a pedestrian's course. The energy function E _i (v _i ′) is expressed by the above equation (A).

In the above equation (A), the parameters of the weighting coefficients λ _{0 to} λ ₂ may be determined empirically by the algorithm designer, but can also be obtained by learning from the actual pedestrian movement. In particular, when using a stationary camera, the environment is constant, and the pedestrian's behavior is assumed to be equivalent to past observation results, so that the effect of learning is great.

  When determining a parameter by learning, captured images of a plurality of pedestrians are acquired by an installation camera. As an example, FIG. 4 is obtained by photographing the vicinity of the lower building entrance from the upper floor window of the building, and shows a plurality of pedestrians. In the captured image obtained by the installation camera, the position of each pedestrian is specified using a method such as manual operation or background difference, and movement information of each pedestrian is acquired. 4 indicates the current position of each pedestrian P, and the line in FIG. 4 indicates the movement trajectory of each pedestrian P. Parameters are learned based on the movement information of each pedestrian P acquired in this way.

Specifically, the first one of the pedestrian (a pedestrian P _A), to predict the position after t seconds using a course prediction model described above. At this time, an arbitrary value is set as the parameter. The actual movement of the pedestrian P _A because it has been found as described above, calculates the distance difference between the actual position of the pedestrian P _A and the position of the pedestrian P _A predicted in any parameter. A value such as the square of the distance is used as this distance difference.

For another pedestrian, the distance difference between the predicted position and the actual position is calculated in the same manner. In this case, as a parameter value, use the same as used when the distance difference calculation for the pedestrian P _A. In this way, the same parameter value is used for all pedestrians, and the respective distance differences are obtained. And the sum of the distance difference about all the pedestrians is calculated | required.

  Here, when the value of the parameter to be used is changed, the predicted position of the pedestrian changes, and accordingly, the distance difference between the predicted position and the actual position also changes. For this reason, the total distance difference for all pedestrians also changes. Therefore, the parameter value is repeatedly corrected using a method such as the steepest descent method so that the total distance difference becomes small. Then, when the total distance difference does not change, it is decided to adopt the parameter value at that time.

  Finding an optimal combination of a plurality of parameters becomes difficult as the number of parameters increases. Therefore, for example, using a genetic algorithm, a set of parameters that are set arbitrarily is evaluated and calculated at the same time, and the parameters are set arbitrarily (mutations in the genetic algorithm) from time to time, so that the optimal parameters You may ask for.

  Since the parameters determined in this way are set based on actual pedestrian behavior, the accuracy of prediction is high, and it is possible to contribute to improving the accuracy of pedestrian tracking.

In addition, when the camera is mounted on a vehicle, the camera moves and cannot be learned in advance. This case can be dealt with by appropriately switching the parameter values according to the nature of the place. The nature of the place can be obtained from a position information acquisition device such as GPS and map information. For example, when the place is on a pedestrian crossing, pedestrians usually move with the end of the pedestrian crossing as their destination, so the restriction of moving to the destination by increasing λ ₂ is limited. Predict the course with a stronger model. In a very congested situation, since allowable inevitable even if the distance is short of pedestrians each other to reduce the lambda _0, the model predicts the course in weakening the limitation of the distance of the pedestrian between.

The total discomfort degree E _i (v _i ′) in the above equation (A) obtained as described above changes depending on how the pedestrian P _i takes the next step. Here, as described above, the next step of the pedestrian P _i is predicted by obtaining the speed vector v _i ′ that minimizes the total number of discomfort degrees E _i (v _i ′).

FIG. 5 shows an example of the number of discomfort levels when the speed vector v _i ′ of the next step of the pedestrian P _i is changed. In FIG. 5, P ₁ and P ₂ indicate pedestrians, respectively. A white point A ₁ indicates a speed vector v _i ′ of the next step when the pedestrian P ₁ proceeds while maintaining the current speed and direction, and a white point A ₂ indicates that the pedestrian P ₂ has the current speed. In addition, the velocity vector v _i ′ of the next step when the vehicle proceeds while maintaining the direction is shown.

Here, 60 degrees on the left and right with respect to the direction indicated by the points A ₁ and A ₂ are changed in increments of 10 degrees, and further changed by 0.3 m / s before and after the speed indicated by the points A ₁ and A ₂ . Indicates the number of discomfort. At this time, the degree of discomfort is lower in the order of the black portion, hatched portion, dot portion, and white portion.

For the pedestrian P ₁ , the discomfort is the lowest at the speed vector v _i ′ indicated by the black point B ₁ that satisfies the conditions of the above items (1) to (3), so that the speed vector v _i ′ is the next step. Selected as. Pedestrian P ₂ is likewise above (1) to (3) _'for unpleasantness in is lowest, its velocity vector v _i' velocity vector v _i shown by black dots B ₂ that satisfies the conditions of the terms following Selected as a step.

  The predicted pedestrian course model described above is a model that can be applied to various behavioral characteristics of pedestrians, and predicts the actual movement of pedestrians more accurately than the predicted course model of vehicles with many restrictions. It is possible. For example, when a large number of pedestrians are crowded, it is predicted that the pedestrians will keep a certain distance while approaching each other. When there are few pedestrians, the pedestrians will head toward the destination. It is predicted to proceed in the shortest distance.

In addition, by adjusting the values of the parameters λ _{0 to} λ ₂ , energy for maintaining a constant distance between pedestrians, energy for maintaining the walking speed of pedestrians, energy for the pedestrians to travel to the destination at the shortest distance It is possible to adjust which one of them is emphasized, and it is possible to flexibly construct a predicted course model that reflects the characteristics of the place and the group.

  Furthermore, by using the predicted course model described above, the pedestrian's course is predicted at an arbitrary future time by repeating the prediction of the next step of the pedestrian by the designated step, and the pedestrian is tracked. Can do.

  FIG. 6 is a diagram illustrating an example of a movement path of a pedestrian detected by image recognition. As a pedestrian detection method, for example, HOG (Histograms of Oriented Gradients) or SVM (Support Vector Machine) is used.

6, when the pedestrian is detected at the position of the point Q ₁ in a certain image frame, from the detection position Q ₀ and the current detection position to Q ₁ the previous image frame, at the current detection position Q ₁ The velocity vector is obtained. When the predicted path of a pedestrian based on the velocity vector by the method described above, and is expected to pedestrian comes to a position Q ₂ at the next time step.

At this time, even in consideration of the prediction error, assuming that there is a pedestrian from the position Q ₂ within a predetermined range R, to perform the image recognition processing only within the range R. As a result, it is possible to reduce the detection of irrelevant background as a pedestrian (false detection). Note that the range R is not limited to a rectangle as illustrated, but may be a circle, an ellipse, or the like.

Further, when the value is close to the HOG features used for recognition in HOG features and position Q ₂ to which was used in the recognition in the position Q ₁ is a pedestrian position Q _1, pedestrian position Q ₂ Can be considered the same. Therefore, in reality, there are pedestrians moving along the route of Q ₀ → Q ₁ → Q ₂ → Q ₃ → Q ₄ and pedestrians moving along the route of Q ₁₀ → Q ₁₁ → Q ₁₂ → Q ₁₃ → Q _14. When passing close, one pedestrian moves along the route Q ₀ → Q ₁ → Q ₂ → Q ₁₃ → Q ₁₄ and the other pedestrian Q ₁₀ → Q ₁₁ → Q ₁₂ → Q ₃ → It is prevented from resulting in mistaken as to move in a path Q _4.

  By using the predicted course model described above, a plurality of pedestrians can be tracked accurately. At this time, the pedestrian can be tracked more accurately as the probability of the pedestrian's course prediction increases.

  Returning to FIG. 1, the pedestrian movement estimation unit 6 predicts the path of the pedestrian using the predicted path model described above, and tracks the pedestrian. The procedure of the pedestrian movement estimation process by the pedestrian movement estimation unit 6 is shown in FIG.

  In FIG. 7, first, the image data obtained by the image processing unit 4 is input (step S11). Subsequently, all pedestrians are detected from the image data (step S12). Then, the position and speed vector of each pedestrian after t seconds are obtained from the current position and speed vector of each pedestrian (step S13).

Next, an energy term I _i relating to the distance between a certain pedestrian (referred to as a specific pedestrian) and other pedestrians existing around the specific pedestrian is obtained (step S14). Also, determine the energy term _{S i} about walking speed of a particular pedestrian (Step S15). Moreover, determining the energy term _{D i} related to destinations of a particular pedestrian (Step S16). Each energy term _{I i,} by using the S i, _{D i} and the storage unit 5 weight coefficient lambda ₀ is stored in the to [lambda] _2, obtains the energy function _{E i} for a particular pedestrian (Step S17).

Next, the course of the specific pedestrian is predicted based on the energy function E _i (step S18). Specifically, a speed vector having a minimum energy function E _i among the speed vectors after t seconds of the specific pedestrian is set as the next step of the specific pedestrian. And a pedestrian is tracked based on a pedestrian's course prediction result (step S19).

  In the above, steps S11 to S13 of the camera 2 and the image processing unit 4 of the ECU 3 and the pedestrian movement estimation unit 6 are movement information for acquiring movement information of the pedestrian and movement information of the moving body existing around the pedestrian. Configure the acquisition means. Step S14 of the pedestrian movement estimation unit 6 constitutes distance related information acquisition means for acquiring information related to the distance between the pedestrian and the moving body based on the movement information of the pedestrian and the movement information of the moving body. The storage unit 5 of the ECU 3 and the steps S17 to S19 of the pedestrian movement estimation unit 6 constitute estimation means for estimating the movement behavior of the pedestrian using information on the distance between the pedestrian and the moving body.

  Moreover, step S15 of the pedestrian movement estimation part 6 comprises the speed related information acquisition means which acquires the information regarding the walking speed of a pedestrian. Step S16 of the pedestrian movement estimation unit 6 constitutes destination related information acquisition means for acquiring information related to the destination of the pedestrian.

As described above, in the present embodiment, the pedestrian uses the energy term I _i considering the property of keeping the distance from other pedestrians at a certain level or higher, and represents the pedestrian's total discomfort degree energy. Since the function E _i is obtained and the path of the pedestrian is predicted based on the energy function E _i , for example, even when the pedestrian moves suddenly away from another pedestrian, the path of the pedestrian is appropriately set. Can be predicted.

In addition, when predicting the path of a pedestrian, in addition to such an energy term I _i , the energy term S _i taking into account the property that the pedestrian wants to maintain a constant speed and the pedestrian must be in a specific destination. Since the energy term D _i taking into account the property of traveling toward the vehicle is also used, the path of the pedestrian can be predicted more appropriately.

  Thereby, it becomes possible to estimate the movement behavior of the pedestrian with high accuracy and improve the tracking performance of the pedestrian.

  The present invention is not limited to the above embodiment. For example, in the above embodiment, information on the distance between the pedestrian and another pedestrian is acquired, but not only information on the distance between the pedestrians, but also information on the distance between the pedestrian and a moving body such as a bicycle. You may acquire and estimate the movement action of a pedestrian.

  Moreover, in the said embodiment, although the pedestrian is detected from the captured image acquired with the camera 2, you may detect a pedestrian other than this by sensors, such as a laser radar and a millimeter wave radar.

DESCRIPTION OF SYMBOLS 1 ... Pedestrian movement estimation apparatus, 2 ... Camera (movement information acquisition means), 3 ... ECU, 4 ... Image processing part (movement information acquisition means), 5 ... Memory | storage part (estimation means), 6 ... Pedestrian movement estimation part (Movement information acquisition means, distance related information acquisition means, speed related information acquisition means, destination related information acquisition means, estimation means).

Claims (2)

Movement information acquisition means for acquiring movement information of a pedestrian and movement information of at least one moving body existing around the pedestrian,
Based on the movement information of the pedestrian and the movement information of the moving body, distance-related information acquisition means for acquiring information on the distance between the pedestrian and the moving body;
Speed-related information acquisition means for acquiring information on the walking speed of the pedestrian;
Destination related information acquisition means for acquiring information about the destination of the pedestrian;
Using information on the distance between the pedestrian and the moving body and information on the destination of the pedestrian, and estimating means for estimating the movement behavior of the pedestrian ,
The estimation means uses a model that defines the behavior of the pedestrian based on information on the distance between the pedestrian and the moving body, information on the destination of the pedestrian, and information on the walking speed of the pedestrian. A pedestrian movement estimation device for estimating movement behavior of the pedestrian. A pedestrian movement estimation method by a pedestrian movement estimation device,
The movement information acquisition means of the pedestrian movement estimation device acquires the movement information of the pedestrian and the movement information of at least one moving body existing around the pedestrian,
The distance-related information acquisition means of the pedestrian movement estimation device acquires information on the distance between the pedestrian and the moving body based on the movement information of the pedestrian and the movement information of the moving body;
The speed-related information acquisition means of the pedestrian movement estimation device acquires information about the walking speed of the pedestrian;
The destination-related information acquisition means of the pedestrian movement estimation device acquires information about the destination of the pedestrian;
The estimation means of the pedestrian movement estimation device estimates the movement behavior of the pedestrian using information on the distance between the pedestrian and the moving body and information on the destination of the pedestrian. Including
In the step of estimating the movement behavior of the pedestrian, the estimation means is based on information on a distance between the pedestrian and the moving body, information on the destination of the pedestrian, and information on the walking speed of the pedestrian. The pedestrian movement estimation method characterized by estimating the movement action of the said pedestrian using the model which prescribes | regulates the said pedestrian's action .

Images