JP7356961B2 - Pedestrian road crossing simulation device, pedestrian road crossing simulation method, and pedestrian road crossing simulation program - Google Patents

Description

The present invention relates to a pedestrian road crossing simulation device, a pedestrian road crossing simulation method, and a pedestrian road crossing simulation program, and particularly relates to a device, method, and program for simulating pedestrian road crossing behavior.

BACKGROUND ART A simulation method has been proposed for understanding the mutual behavior of a pedestrian attempting to cross a road and a vehicle traveling on the road (see, for example, Patent Document 1). In Patent Document 1, a virtual pedestrian is automatically generated based on route specifications and behavior specifications, and a digital simulation including a virtual pedestrian crossing an intersection and a virtual vehicle reacting to the virtual pedestrian crossing the intersection is executed. The configuration is disclosed. In the same document, the behavior specification data 195 shows that the virtual pedestrian increases its walking speed as the crosswalk timer approaches zero, or that the virtual pedestrian modifies its moving speed or direction in response to the approach of a virtual vehicle ( For example, it is disclosed that the user quickly moves away from an approaching virtual vehicle while increasing the moving speed.

In particular, when verifying the driving control system installed in self-driving vehicles, which are currently being developed, we use a simulation (crossing decision model) to understand the mutual behavior of pedestrians attempting to cross the road and vehicles traveling on the road. ) is required.

The crossing decision model of the current simulation method is limited to a binary decision (single decision model) in which a virtual pedestrian in the simulation environment either crosses or does not cross the road in the simulation environment. Therefore, in the case of the conventional single decision model, it is not possible to make complex decisions that take into account the speed (walking, running) of the virtual pedestrian when crossing the street. Furthermore, with a single decision model, it is complicated and difficult to express changes in pedestrian behavior in response to the movement of a running vehicle. For this reason, when trying to make complex judgments about the relationship between pedestrians and running vehicles, the difficulty of modeling becomes high.

Therefore, with regard to the relationship between pedestrians and vehicles, there is a strong need to develop a judgment model that takes into account the mutual relationship, avoids complication in walking judgment, and approaches the behavior of actual pedestrians more closely.

JP2019-109881A

The present invention has been made in view of the above-mentioned points, and with regard to the relationship between a pedestrian and a running vehicle in a simulation environment, a crossing decision is generated using the walking speed of the pedestrian, which complicates the pedestrian's walking decision. The present invention provides a pedestrian road crossing simulation device that realizes a walking judgment model that avoids the behavior of pedestrians and approaches actual pedestrian behavior, and also provides a pedestrian road crossing simulation method and a pedestrian road crossing simulation program.

That is, the pedestrian road crossing simulation device of the embodiment includes a walking information reception unit that receives walking information related to the walking of a virtual pedestrian crossing the road within the simulation environment, and a walking information reception unit that receives walking information related to the walking of a virtual pedestrian crossing the road within the simulation environment. A driving information reception unit that receives driving information related to driving, and a driving information reception unit that receives driving information related to driving, and a driving information reception unit that receives driving information related to driving, and a virtual pedestrian whose walking information is inputted crosses or does not cross a road in the simulation environment, depending on the driving of the virtual vehicle into which the driving information is inputted. The present invention is characterized by comprising: a crossing judgment section that generates a crossing judgment that defines the behavior of a virtual pedestrian; and an action determining section that outputs a crossing judgment of the virtual pedestrian.

Furthermore, it may include a virtual pedestrian generation section that generates a virtual pedestrian within the simulation environment, and a virtual vehicle generation section that generates a virtual vehicle within the simulation environment.

Further, the travel information may include at least the speed of the virtual vehicle on the road and the distance between the virtual vehicle and the virtual pedestrian. Furthermore, the walking information may include the walking speed of the virtual pedestrian on the road.

Furthermore, when the crossing judgment unit generates the crossing judgment, a plurality of judgment models may be presented according to walking information of the virtual pedestrian.

Furthermore, the crossing judgment section is provided with a judgment model selection section that selects a plurality of judgment models, and the judgment model selection section selects one judgment model from the plurality of judgment models according to walking information of the virtual pedestrian. good.

Further, the judgment model selection unit may select one judgment model from a plurality of judgment models according to the road crossing speed of the virtual pedestrian as the walking information of the virtual pedestrian.

Furthermore, the cross-cutting judgment may be generated by logistic regression analysis.

The pedestrian road crossing simulation device of the present invention includes a walking information reception unit that receives walking information related to the walking of a virtual pedestrian crossing the road in a simulation environment, and a walking information reception unit that receives walking information related to the walking of a virtual pedestrian crossing the road in the simulation environment. A driving information reception unit that receives related driving information, and a virtual machine that allows a virtual pedestrian whose walking information is input to cross or not cross a road in the simulation environment according to the driving of a virtual vehicle whose driving information is input. Since it is equipped with a crossing judgment unit that generates a crossing judgment that defines the behavior of a pedestrian, and an action determining unit that outputs a crossing judgment of a virtual pedestrian, the pedestrian The pedestrian's walking speed is used to generate the crossing decision, which avoids complicating the pedestrian's walking decision and makes it possible to simulate a walking decision model that more closely resembles the behavior of actual pedestrians. Further, similar simulation of the walking judgment model can be realized in a pedestrian road crossing simulation method and a pedestrian road crossing simulation program.

FIG. 1 is a block diagram showing an example of the configuration of a pedestrian road crossing simulation device according to an embodiment. FIG. 2 is a block diagram showing the configuration of functional units of the pedestrian road crossing simulation device. It is a block diagram showing the flow of processing of the pedestrian road crossing simulation device. It is a flowchart which shows the process of cross-cutting judgment. FIG. 3 is a schematic diagram showing selection of a judgment model. It is a table showing an example of cross-cutting judgment in the action determining unit. It is a 1st flowchart which shows the main control of a pedestrian road crossing simulation apparatus. It is a 2nd flowchart which shows the main control of a pedestrian road crossing simulation apparatus. It is a 3rd flowchart which shows the main control of a pedestrian road crossing simulation apparatus.

The pedestrian road crossing simulation device of the embodiment generates virtual pedestrians, virtual roads, virtual vehicles traveling on the roads, etc. as data within the simulation environment of the device. Then, by appropriately changing conditions such as the walking speed of the virtual pedestrian and the speed of the virtual vehicle, the judgment of whether or not the virtual pedestrian crosses the road is determined under various conditions in the simulation environment. . Crossing decisions determined in such a simulation environment can be used, for example, to improve the accuracy of vehicle travel control and develop advanced driver-assistance systems (ADAS) to realize unmanned vehicle operation. be done.

The block diagram in FIG. 1 shows the configuration of main parts of a pedestrian road crossing simulation device 1 according to an embodiment. The pedestrian road crossing simulation device 1 includes a processing section 10 (computer) that executes calculations related to simulation, and a display section 20 connected to the processing section 10.

The processing unit 10 includes hardware such as a CPU 11, a ROM 12, a RAM 13, a storage unit 14, and an I/O (in/out interface) 15. In addition, main memory, LSI, etc. are also included. Also, in terms of software, it is realized by a synchronization program etc. loaded into the main memory.

When each functional unit of the processing unit 10 is implemented by software, the processing unit 10 is implemented by executing instructions of a program that is software that implements each function. A "non-temporary tangible medium" such as a CD, a DVD, a semiconductor memory, a programmable logic circuit, etc. can be used as a recording medium for storing this program. Further, this program may be supplied to the processing unit 10 (computer) via any transmission medium (communication network, broadcast waves, etc.) that can transmit the program.

The processing unit 10 is a variety of electronic computers (calculation resources, computers) such as a personal computer (PC), a mainframe, a workstation, a cloud computing system, a smartphone, a tablet terminal, and the like.

The display unit 20 is a known display or the like. In addition, when the processing section 10 is a smartphone, a tablet terminal, etc., the display section 20 may be included in the processing section 10.

The state of the simulation environment of the pedestrian road crossing simulation device 1 is schematically shown as a display image 50 on the display unit 20 in FIG. A virtual pedestrian H is shown in a display image 50 showing a simulation environment. A road 51 and a crosswalk 52 are displayed within the simulation environment. A virtual vehicle 55 is running toward a crosswalk 52. Under this simulation environment, the pedestrian road judgment simulation device 1 determines whether the virtual pedestrian H crosses the crosswalk 52 or not, based on information given to the virtual pedestrian H (walking information described below). ) to determine.

Various storage units in the processing unit 10 include a ROM 12 and a RAM 13, and the storage unit 14 is a known storage device (not shown) such as an HDD or an SSD. Each functional unit that executes arithmetic processing is an arithmetic element such as the CPU 11.

An I/O (in/out interface) 15 is an interface, buffer, etc. for communication (transmission/reception). The I/O 15 is used to transmit and receive data to and from the display unit 20 for input and output. Further, appropriate operating devices such as a keyboard and a mouse are also connected to the I/O 15 as necessary.

Each functional unit in the CPU 11 of the processing unit 10 is shown in the schematic block diagram of FIG. Each functional unit includes a virtual pedestrian generation unit 110, a virtual vehicle generation unit 120, a walking information reception unit 130, a travel information reception unit 140, a crossing determination unit 150, a behavior determination unit 160, a determination model selection unit 170, and the like.

An overview of the processing flow of the pedestrian road crossing simulation device 1 is shown as a block diagram in FIG. First, input variables are calculated (S1). The input variables are walking information and running information, which will be described later. Next, a crossing judgment is made (S2). The crossing judgment is a judgment of the virtual pedestrian H in the simulation environment as to whether or not to cross the crosswalk 52. Then, an action decision is made (S3). In the action determination, a decision is made to stop the virtual pedestrian H in the simulation environment from crossing the crosswalk 52, and furthermore, a crossing manner such as walking or running is determined.

The virtual pedestrian generation unit 110 generates a virtual pedestrian crossing a road within the simulation environment. Specifically, the virtual pedestrian generation unit 110 generates a virtual pedestrian H in the display image 50 of FIG. 1 . The number of virtual pedestrians H is expanded from one to a plurality of people according to walking information to be described later. The aspect of the virtual pedestrian H in the display image 50 of FIG. 1 is a display for convenience of explanation, and may be taken as intangible data. Note that the virtual pedestrian H may be generated by the pedestrian road crossing simulation device 1 itself (virtual pedestrian generation unit 110), or may be set from outside.

The virtual vehicle generation unit 120 generates a virtual vehicle that travels on a road within the simulation environment. Specifically, the virtual vehicle generation unit 120 generates a virtual vehicle like the virtual vehicle 55 in the display image 50 of FIG. The number of virtual vehicles 55 within the simulation environment may be one or more such that they travel sequentially. The aspect of the virtual vehicle 55 in the display image 50 in FIG. 1 is also a display for convenience of explanation, and may be taken as intangible data. Note that the virtual vehicle 55 may be generated by the pedestrian road crossing simulation device 1 itself (virtual vehicle generation unit 120), or may be set from outside.

The walking information receiving unit 130 receives walking information related to the walking of the virtual pedestrian H crossing the road within the simulation environment. In the walking information reception unit 130, the walking information related to the walking of the virtual pedestrian H is the walking speed of the virtual pedestrian H, and is the crossing speed when the virtual pedestrian H crosses the crosswalk 52. The walking information may be, for example, speeds corresponding to the three preset stages of "walking," "running," and "sprinting," or may be expressed as speeds of 1 m/s and 2 m/s (s is seconds). ). Input of walking information for these virtual pedestrians H is accepted (see S1 in FIG. 3).

The driving information receiving unit 140 receives driving information related to the driving of the virtual vehicle 55 passing on the road 51 within the simulation environment. The traveling information related to the traveling of the virtual vehicle 55 includes the vehicle speed (speed) (m/s), vehicle acceleration (m/s ² ), etc. of the virtual vehicle 55. Furthermore, the distance (m) between the virtual pedestrian H and the virtual vehicle 55 (pedestrian-vehicle distance) is also added to the traveling information. Input of travel information for these virtual vehicles 55 is accepted (see S1 in FIG. 3).

Further, the distance of the crosswalk 52 (width of the road 51) (m), the inclination angle (°) of the road 51, etc. that the virtual pedestrian H in the simulation environment attempts to cross may be added to the travel information (Fig. (See S1 of 3).

The crossing determination unit 150 determines whether the virtual pedestrian H whose walking information has been input crosses or does not cross the road 51 (crosswalk 52) in the simulation environment, according to the travel of the virtual vehicle 55 whose travel information has been input. A crossing judgment is generated that defines the behavior of the virtual pedestrian H based on the walking information.

Then, the behavior determining unit 160 outputs the crossing judgment of the virtual pedestrian H. The result of the crossing determination (virtual pedestrian H crosses by walking. Virtual pedestrian H crosses by running. Virtual pedestrian H stays) is output to the display unit 20.

The flowchart of FIG. 4 is shown as an example of the process of cross-cutting determination. As described above, walking information and running information are input as input variable calculations (S11). At this time, it is determined from the walking information and travel information whether it is possible to cross the crossing on foot (S12). If it is possible to cross the street on foot, a crossing determination that it is possible to cross on foot is generated for the virtual pedestrian H (S14). If it is determined in S12 that it is not possible to cross on foot, it is determined whether it is possible to cross on foot (S13). If it is possible to cross by running, a crossing determination that it is possible to cross by running is generated for the virtual pedestrian H (S15). Note that if it is determined in S13 that it is not possible to cross by running, a crossing determination of not crossing (stopping) is generated (S16).

The walking or running branch in the flowchart of FIG. 4 is a simplified model, and for example, a trotting branch or the like may be further included between walking and running.

When the crossing judgment unit 150 generates the crossing judgment, a plurality of judgment models are presented in advance according to the walking information of the virtual pedestrian H. Therefore, the judgment model selection unit 170 selects the optimum judgment model based on the input conditions from among the plurality of judgment models. A judgment model is a model that performs calculations to arrive at a conclusion (cross the line or not) based on predetermined numerical calculations based on input items.

When selecting a judgment model by the judgment model selection unit 170, the schematic diagram of FIG. 5 is referred to. First, when the virtual pedestrian H crosses the crosswalk 52, two or more steps of crossing speed [v _p (m/s)] of "walking" and "running" are input as walking information. For simplicity, "walking" and "running" are shown in the diagram. In the example of FIG. 5, the distance between the virtual pedestrian H and the virtual vehicle 55 (pedestrian-to-vehicle distance) [d(m)], the vehicle speed of the virtual vehicle 55 [v _c (m/s)], and the vehicle Acceleration [α _c (m/s ² )] is input.

Therefore, the optimal judgment model is selected based on the input items mentioned above. In the example of FIG. 5, "judgment model I" or "judgment model II" is selected from a plurality of determination models according to the crossing speed [v _p (m/s)] of the virtual pedestrian H (in FIG. (See S2).

If the crossing speed [v _p (m/s)] of the virtual pedestrian H is the walking speed, "determination model I" is selected, and if the crossing speed [v _p (m/s)] is the running speed "Judgment Model II" is selected. It is more convenient to classify and select judgment models based on the crossing speed of the virtual pedestrian H.

Then, the judgment results of the individual judgment models are compared and output as a final conclusion. The number of decision models to be set is not limited to two types as in the example of FIG. For example, depending on the crossing speed of the virtual pedestrian H, a judgment model corresponding to brisk walking between walking and running may be set, or a judgment model with five levels depending on the crossing speed may be set.

According to the pedestrian road crossing simulation device 1 of the embodiment, a logistic regression analysis model is used as a judgment model for crossing judgment. Crossing decisions are calculated using a logistic regression analysis model, and the crossing probability is compared with an arbitrary threshold value. If the crossing probability exceeds the threshold, "go" (cross), and if it is below the threshold, "wait" (wait). The conclusion is output. The calculation of the crossing probability and the results of the crossing judgment are obtained by equations (i), (ii), and (iii).

In the formula, "result" is the result of the cross-cutting judgment, and "thresh" is an arbitrary threshold value. "p" is the crossing probability, "k" is the variable for calculating the crossing probability, "v _c " is the vehicle speed, "α _c " is the vehicle acceleration, "d" is the distance between pedestrians and vehicles, and "v _p " is the walking speed ( traverse speed), and “η _i ” indicates the regression coefficient in the logistic regression.

The model of the logistic regression analysis used is a model that uses walking speed as an input variable, and by inputting different walking speeds to the same model, it is possible to obtain an arbitrary pattern of crossing judgment results. For example, assuming that the walking speed of the virtual pedestrian H is 1 m/s when walking and 3 m/s when running, the cross-crossing determination result (crossing/waiting) for each case is calculated.

Based on the crossing judgment results for walking and running obtained by the logistic regression model of equations (i), (ii), and (iii) above, crossing behavior is determined according to the table in Figure 6 ( (See S3 in FIG. 3). The output of the final crossing judgment result is either "walk" (walking across: 1 m/s), "run" (running across: 3 m/s), or "wait" (wait, do not cross). Become. In the table, cases where it is possible to cross by walking but not by running are generally considered unlikely to occur and are marked with a "-" (excluded from consideration).

The function of equation (ii) is a type of link function, and in the logistic regression model of the embodiment, the crossing probability is clearly calculated based on the input value, and the crossing probability is classified using a threshold value. In addition to the logistic regression model of the embodiment, a known deep learning method may be used to determine the probability of crossing and the action of determining the crossing.

As in the embodiment, a judgment model to be applied is first selected in advance based on a simple classification (walking speed), a crossing judgment is calculated using the selected judgment model, and a judgment result is output. In this way, by classifying and selecting a decision model, calculations are performed on the decision model to be used to derive a cross-cutting probability, that is, a behavioral decision for a cross-cutting decision. Therefore, depending on the crossing speed of the virtual pedestrian, it is possible to individually determine the crossing decision behavior, and a walking decision model that approaches the behavior of an actual pedestrian is realized.

Next, a pedestrian road crossing simulation method according to an embodiment will be explained together with a pedestrian road crossing simulation program. The pedestrian road crossing simulation method of the embodiment is executed by the CPU 11 of the processing unit 10 based on a pedestrian road crossing simulation program. The pedestrian road crossing simulation method causes the CPU 11 of the processing unit 10 to execute a virtual pedestrian generation function, a virtual vehicle generation function, a walking information reception function, a driving information reception function, a crossing determination function, and an action determination function, and further includes: Execute the judgment model selection function. Since each function overlaps with the above explanation, the details will be omitted.

The flowcharts in FIGS. 7, 8, and 9 are the overall flow of the pedestrian road crossing simulation method in the CPU 11 of the processing unit 10, and in FIG. 7, a virtual pedestrian generation step (S110), a virtual vehicle generation step (S120), In FIG. 8, a walking information reception step (S130), a travel information reception step (S140), a crossing determination step (S150), and an action determination step (S160) are executed, and in FIG. 9, a determination model selection step (S170) is executed. .

In the CPU 11 of the processing unit 10, the processing in the virtual pedestrian generation step (S110) corresponds to the description of the virtual pedestrian generation unit 110. The processing in the virtual vehicle generation step (S120) corresponds to the description of the virtual vehicle generation unit 120. The process in the walking information receiving step (S130) corresponds to the description of the walking information receiving unit 130. The processing in the driving information receiving step (S140) corresponds to the description of the driving information receiving unit 140. The processing in the crossing judgment step (S150) corresponds to the description of the crossing judgment unit 150. The processing in the action determining step (S160) corresponds to the description of the action determining unit 160. Further, the processing in the judgment model selection step (S170) corresponds to the description of the judgment model selection unit 170.

The pedestrian road crossing simulation device, pedestrian road crossing simulation method, and pedestrian road crossing simulation program of the present invention avoid complicating walking decisions while taking into account the mutual relationship between pedestrians and running vehicles within the simulation environment. , we have realized a walking judgment model that more closely resembles the behavior of actual pedestrians. For this reason, it is a promising alternative to existing pedestrian road crossing simulation devices.

1 Pedestrian road crossing simulation device 10 Processing unit (computer)
11 CPU
12 ROM
13 RAM
14 Storage section 15 I/O (in/out interface)
20 Display unit 50 Display image 51 Road 52 Crosswalk 55 Virtual vehicle H Virtual pedestrian 110 Virtual pedestrian generation unit 120 Virtual vehicle generation unit 130 Walking information reception unit 140 Traveling information reception unit 150 Crossing determination unit 160 Behavior determination unit 170 Judgment model selection section

Claims (10)

A pedestrian road crossing simulation device that simulates pedestrian road crossing behavior,
a walking information reception unit that receives walking information related to walking of a virtual pedestrian crossing a road in a simulation environment;
a driving information reception unit that receives driving information related to driving of a virtual vehicle passing on the road in the simulation environment;
In accordance with the travel of the virtual vehicle to which the travel information has been input, the virtual pedestrian to which the walking information has been input crosses or does not cross the road in the simulation environment; , a crossing determination unit that generates a prescribed crossing determination based on the walking information;
A pedestrian road crossing simulation device comprising: an action determining unit that outputs the crossing judgment of the virtual pedestrian. a virtual pedestrian generation unit that generates the virtual pedestrian within the simulation environment;
The pedestrian road crossing simulation device according to claim 1, further comprising a virtual vehicle generation unit that generates the virtual vehicle within the simulation environment. The pedestrian road crossing simulation device according to claim 1 or 2, wherein the travel information includes at least a vehicle speed of the virtual vehicle on the road and a distance between the virtual vehicle and the virtual pedestrian. The pedestrian road crossing simulation device according to any one of claims 1 to 3, wherein the walking information includes a walking speed of the virtual pedestrian on the road. When generating the crossing judgment in the crossing judgment unit, a plurality of judgment models are presented for judging whether or not the virtual pedestrian crosses the road based on numerical calculation according to the walking information of the virtual pedestrian. A pedestrian road crossing simulation device according to any one of claims 1 to 4. The cross-sectional judgment unit includes a judgment model selection unit that selects the plurality of judgment models,
The pedestrian road crossing simulation device according to claim 5, wherein the judgment model selection unit selects one judgment model from the plurality of judgment models according to the walking information of the virtual pedestrian. The pedestrian according to claim 6, wherein the judgment model selection unit selects one judgment model from the plurality of judgment models as the walking information of the virtual pedestrian according to the cross speed of the virtual pedestrian on the road. Road crossing simulation device. The crossing judgment includes a crossing probability that is a probability that the virtual pedestrian will cross, the vehicle speed of the virtual vehicle, the vehicle acceleration of the virtual vehicle, the distance between the virtual pedestrian and the virtual vehicle, and the walking of the virtual pedestrian. The pedestrian road crossing simulation device according to any one of claims 1 to 7, wherein the pedestrian road crossing simulation device is generated by a logistic regression analysis using a walking speed that is a speed of . A pedestrian road crossing simulation method for simulating pedestrian road crossing behavior, the method comprising:
The computer is
a walking information receiving step for receiving walking information related to walking of a virtual pedestrian crossing a road in a simulation environment;
a driving information receiving step of receiving driving information related to driving of a virtual vehicle passing on the road in the simulation environment;
In accordance with the travel of the virtual vehicle to which the travel information has been input, the virtual pedestrian to which the walking information has been input crosses or does not cross the road in the simulation environment; a cross-decision step for generating a prescribed cross-decision;
A pedestrian road crossing simulation method, comprising: executing a behavior determining step of outputting the crossing judgment of the virtual pedestrian. A pedestrian road crossing simulation program device for simulating pedestrian road crossing behavior,
The computer is
a walking information reception function that receives walking information related to the walking of a virtual pedestrian crossing a road in a simulation environment;
a driving information reception function that receives driving information related to driving of a virtual vehicle passing on the road in the simulation environment;
In accordance with the travel of the virtual vehicle to which the travel information has been input, the virtual pedestrian to which the walking information has been input crosses or does not cross the road in the simulation environment; a cross-determination function that generates a prescribed cross-determination;
A pedestrian road crossing simulation program comprising: an action determining function that outputs the crossing judgment of the virtual pedestrian.

Images