Vehicle crossing estimation system

Abstract

To provide a crossing estimation system for a vehicle capable of accurately estimating whether or not a pedestrian intends to cross.SOLUTION: In the vehicle crossing intention estimation system 10, the gate setting unit 22C sets the crosswalk gate GX for the crosswalk C as one of the plurality of exit gates Gk, and the probability distribution data PD includes at least the probability distribution Pr (v p) of the speed vectors v of pedestrians moving from the virtual gate GVa to the outside of the virtual area VRa, for each of the plurality of positions p in the virtual area VRa with respect to the virtual gate GVa. The crossing intention estimation unit 22E applies the probability distribution data PD to each of the plurality of exit gates Gk independently to calculate the likelihood Lk that the pedestrian A walking in the vicinity area R passes through each exit gate Gk, and when the exit gate Gk having the maximum likelihood Lk is the crosswalk gate GX, estimates that the pedestrian A in the vicinity area R has the crossing intention to cross the crosswalk C.SELECTED DRAWING: Figure 11

Description

[Technical Field]

[0001] The present invention relates to a vehicle crossing estimation system, and more particularly to a vehicle crossing estimation system that estimates whether a pedestrian near a crosswalk intends to cross the crosswalk. [Background technology]

[0002] Autonomous vehicle driving technology requires logic that estimates the possibility that a pedestrian walking on a sidewalk near a crosswalk without traffic lights will cross the crosswalk to cross the roadway, and that slows down and stops the vehicle in front of the crosswalk based on the estimation result. Conventionally, as a method for such estimation, a method is known that detects or determines the crossing state of a pedestrian based on the position of the pedestrian near the crosswalk and changes in that position (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7359099 Summary of the Invention [Problem to be solved by the invention]

[0004] However, the method described in Patent Document 1 has a problem in that it may erroneously assume that a pedestrian intends to cross a crosswalk simply because they are near the crosswalk, resulting in low estimation accuracy. Therefore, with the conventional technology, such erroneous estimation can lead to the vehicle unnecessarily slowing down or stopping in front of the crosswalk.

[0005] Therefore, the present invention has been made to solve the above-mentioned problems, and aims to provide a vehicle crossing estimation system that can accurately estimate whether or not a pedestrian intends to cross the road. [Means for solving the problem]

[0006] In order to solve the above-mentioned problems, the present invention provides a vehicle crossing estimation system that includes a crosswalk detection unit that detects a crosswalk without traffic lights ahead of a vehicle on a travel lane, a gate setting unit that virtually sets a plurality of exit gates for pedestrians to exit from within a vicinity area of ​​the crosswalk including pedestrians connected to the detected crosswalk, a memory unit that stores probability distribution data including at least a probability distribution of velocity vectors within the virtual area that pedestrians who exit from within the virtual area through the virtual gates have, a pedestrian detection unit that detects pedestrians within the vicinity area of ​​the crosswalk, and a crossing intention estimation unit that estimates whether or not a pedestrian detected within the vicinity area has an intention to cross the crosswalk. The exit gate setting unit sets a crosswalk gate for the crosswalk as one of a plurality of exit gates, and the probability distribution data has, for each of a plurality of positions within the virtual area for the virtual gate, a probability distribution of the velocity vector of a pedestrian exiting the virtual area from the virtual gate, as well as a probability distribution of the pedestrian's velocity vector and the probability that the pedestrian will pass through the virtual gate, and the crossing intention estimation unit applies the probability distribution data independently to each of the plurality of exit gates to calculate the likelihood that a pedestrian walking within the nearby area will pass through each exit gate, and if the exit gate with the greatest likelihood is the crosswalk gate, it estimates that the pedestrian within the nearby area has the intention to cross the crosswalk.

[0007] According to the present invention configured as described above, it is possible to estimate whether a pedestrian intends to cross a crosswalk by referring to the probability distribution data of the velocity vector of the pedestrian near the crosswalk. Therefore, the present invention can estimate a pedestrian's intention to cross a crosswalk with higher accuracy than conventional methods that estimate a pedestrian's intention to cross based only on limited information such as the pedestrian's position and behavior. Furthermore, this makes it possible to avoid vehicles unnecessarily slowing down or stopping at crosswalks without traffic lights.

[0008] In the present invention, preferably, the gate setting unit does not set an exit gate at a fixed object on the ground at the boundary of the neighborhood area that prevents pedestrians from exiting the neighborhood area, but sets at least a crosswalk gate for crossing a crosswalk and an exit gate at the boundary between the sidewalk and the neighborhood area. According to the present invention configured in this way, pedestrians can exit from the neighborhood area to the outside of the neighborhood area through any one of the exit gates.

[0009] In the present invention, the probability distribution data is preferably statistical data based on observation data of pedestrians walking on a predetermined sidewalk connected to a predetermined crosswalk serving as a virtual gate. According to the present invention configured in this manner, a probability distribution of the speed vectors of pedestrians selecting a crosswalk can be set for a nearby area based on actual observation data.

[0010] In the present invention, preferably, when the crossing intention estimation unit estimates that a pedestrian in the vicinity area has an intention to cross, it activates the vehicle's braking system or issues a predetermined notification to the driver in the vehicle. According to the present invention configured in this way, when it is estimated that a pedestrian has an intention to cross at a crosswalk without traffic lights, it is possible to decelerate the vehicle, etc. This makes it possible to avoid unnecessarily decelerating the vehicle when it is not estimated that the pedestrian has an intention to cross.

[0011] In the present invention, preferably, the virtual region of the probability distribution data has a central boundary line and extends from the central boundary line in a predetermined direction perpendicular to the central boundary line, the virtual gate extends from a first endpoint on the central boundary line to a second endpoint a predetermined width away in the predetermined direction, and the crossing intention estimation unit forms another virtual region having a probability distribution that is line-symmetrical to the virtual region with respect to the central boundary line, and applies the probability distribution data of the virtual region and the another virtual region to the exit gate. According to the present invention configured in this manner, the memory unit stores the probability distribution data of only one side of the virtual region (e.g., the left side of the virtual gate) in advance, and when applying it to the exit gate, a separate virtual region with an inverted probability distribution is formed. Therefore, in the present invention, the amount of probability distribution data stored in the memory unit can be reduced.

[0012] In the present invention, preferably, the crossing-intention estimation unit aligns the second end points of the virtual gates in the virtual area and another virtual area so that they coincide with the two end points of the exit gate, and applies the probability distribution data to the exit gate. According to the present invention configured in this way, the pre-stored probability distribution data PD can be commonly used for exit gates facing in various directions and having various length dimensions. [Effects of the Invention]

[0013] The vehicle crossing estimation device of the present invention can accurately estimate the presence or absence of a pedestrian crossing position. [Brief explanation of the drawings]

[0014] [Figure 1] 1 is an illustration of a vehicle crossing estimation system according to an embodiment of the present invention; [Figure 2] 1 is a system configuration diagram of a vehicle crossing estimation system according to an embodiment of the present invention. [Figure 3] FIG. 1 is an explanatory diagram of probability distribution data according to an embodiment of the present invention. [Figure 4] FIG. 2 is an illustration of a pedestrian crossing pattern according to an embodiment of the present invention. [Figure 5] FIG. 2 is an illustration of a pedestrian crossing pattern according to an embodiment of the present invention. [Figure 6] FIG. 1 is a schematic explanatory diagram illustrating the application of probability distribution data according to an embodiment of the present invention. [Figure 7] FIG. 1 is an explanatory diagram of an observation method for acquiring probability distribution data according to an embodiment of the present invention. [Figure 8] 10A and 10B are explanatory diagrams showing analysis results for obtaining probability distribution data according to an embodiment of the present invention. [Figure 9A] FIG. 1 is an explanatory diagram of a method for acquiring probability distribution data according to an embodiment of the present invention. [Figure 9B] FIG. 1 is an explanatory diagram of a method for acquiring probability distribution data according to an embodiment of the present invention. [Figure 9C]FIG. 1 is an explanatory diagram of a method for acquiring probability distribution data according to an embodiment of the present invention. [Figure 10] FIG. 10 is an explanatory diagram of an example of application of probability distribution data according to an embodiment of the present invention. [Figure 11] FIG. 1 is an explanatory diagram of crossing intention estimation according to an embodiment of the present invention. [Figure 12A] 10 is an example of applying probability distribution data to an exit gate according to an embodiment of the present invention. [Figure 12B] 10 is an example of applying probability distribution data to an exit gate according to an embodiment of the present invention. [Figure 12C] 10 is an example of applying probability distribution data to an exit gate according to an embodiment of the present invention. [Figure 13] 10 is an example of an estimation result of crossing intention estimation according to an embodiment of the present invention. [Figure 14] 10 is a flowchart illustrating a process for estimating a crossing intention according to an embodiment of the present invention. [Figure 15] 10 is a flowchart illustrating a process for estimating a crossing intention according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0015] [System Overview] A vehicle crossing estimation system according to an embodiment of the present invention will now be described with reference to the accompanying drawings. First, an overview of the vehicle crossing estimation system of this embodiment will be described with reference to FIG. 1. FIG. 1 is a schematic explanatory diagram of a vehicle crossing intention estimation system 10 of this embodiment. In the example of FIG. 1, pedestrian A is walking on a sidewalk S adjacent to a traffic lane L. Pedestrian A can choose from route a, which crosses crosswalk C; route b, which continues walking on sidewalk S; or route c, which enters a building from entrance B. However, at the time shown in FIG. 1, vehicle 1 generally cannot determine whether pedestrian A intends to cross crosswalk C without a traffic light. In this specification, "crosswalk" refers to a crosswalk without a traffic light.

[0016] Therefore, in the past, in the case of the example in Figure 1, vehicle 1 would slow down or stop before crosswalk C. However, if pedestrian A chooses route a, vehicle 1 needs to slow down or stop before crosswalk C, but if pedestrian A chooses route b or route c, vehicle 1 does not essentially need to slow down or stop.

[0017] Therefore, in this embodiment, as shown in FIG. 1 , when vehicle 1 is traveling on lane L, a vehicle crossing intention estimation system 10 (hereinafter simply referred to as "system 10") mounted on vehicle 1 is configured to estimate whether pedestrian A walking near crosswalk C will cross crosswalk C (i.e., whether or not pedestrian A has the intention to cross). If the system 10 estimates that pedestrian A has the intention to cross, it is configured to automatically activate the braking device of vehicle 1 so as to slow down or stop vehicle 1 before crosswalk C, or to notify the driver of vehicle 1 of this. On the other hand, if the system 10 estimates that pedestrian A does not have the intention to cross, it avoids unnecessary slowing down or stopping of vehicle 1.

[0018] [Overall system configuration] Next, the configuration of a system 10 of this embodiment will be described with reference to Fig. 2. As shown in Fig. 2, the system 10 includes a control device 20, a sensor 30 that detects the vehicle 1 and its surrounding conditions, an actuator 40 that operates the vehicle 1, and a notification device 50 that notifies the driver of the vehicle 1 of predetermined information. The control device 20 is made up of a computer device including a processor 22, a storage unit 24, etc., and is configured to execute automatic driving control of the vehicle 1 based on sensor information acquired by the sensor 30. The storage unit 24 also stores probability distribution data PD for estimating pedestrians' crossing intentions, map data MD that is map information, an operating program, etc.

[0019] The sensors 30 include a camera 31 that captures images outside the vehicle, a radar that detects targets outside the vehicle, various sensors that detect the behavior of the vehicle 1 (vehicle speed sensor, acceleration sensor, yaw rate sensor, steering angle sensor, accelerator sensor, brake sensor, etc.), a positioning system 32 (GPS, gyro sensor, etc.) that detects the vehicle position, and a navigation system that guides the route of the vehicle 1. The camera 31 acquires image data that captures images of the outside of the vehicle 1 from at least the front to the sides.

[0020] The actuator 40 includes a system (for example, a brake control system, an engine control system, and a steering control system) that controls the behavior of the vehicle 1. The notification device 50 includes a display device, a speaker, a lamp, etc. for notifying the driver of the vehicle 1 of predetermined information.

[0021] The processor 22 of the control device 20 executes automatic driving control in accordance with the installed program, and also functions as a vehicle position estimation unit 22A, a crosswalk detection unit 22B, a gate setting unit 22C, a pedestrian detection unit 22D, a crossing intention estimation unit 22E, and a route generation unit 22F.

[0022] The vehicle position estimation unit 22A acquires the vehicle position of the vehicle 1 from sensor information of the positioning system 32 or the like, and estimates the vehicle position on the map data MD. The crosswalk detection unit 22B detects the position of a crosswalk C without traffic lights by performing image analysis on image data captured by the camera 31 or by using the map data MD.

[0023] The gate setting unit 22C performs image analysis on the image data acquired by the camera 31 or uses the map data MD to analyze the structure around the detected crosswalk (including the roadway, sidewalk, buildings, fixed objects on the ground, facilities (parks, etc.), crosswalks, etc.), and virtually sets multiple exit gates (e.g., Ga, Gb, and Gc in FIG. 11 ) at the boundary of the neighborhood area R of the crosswalk C, including the sidewalk S connected to the crosswalk C, to allow pedestrians to exit from the neighborhood area R to the outside of the neighborhood area R. An exit gate (e.g., Gb) is set at the boundary between the sidewalk S and the crosswalk C connected to it. Note that, as shown in FIG. 11 , an exit gate does not necessarily have to be set in a direction in which pedestrian A is estimated not to be traveling (e.g., a rear area), but an exit gate may be set in the direction in which pedestrian A is estimated to be traveling (e.g., from the front area to the side area).

[0024] The pedestrian detection unit 22D detects the position p and velocity vector v (speed and direction) of pedestrian A around the vehicle 1 by performing image analysis of the image data acquired by the camera 31 and / or using detection data by radar. The crossing intention estimation unit 22E estimates whether or not pedestrian A intends to cross the crosswalk C using the detected behavior information of pedestrian A near the crosswalk C (position p, velocity vector v, etc.) and probability distribution data PD, etc.

[0025] The route generation unit 22F generates a driving route (including the position and speed of each point on the route) for the vehicle 1 during autonomous driving control based on sensor information. The route generation unit 22F also generates a generated route for slowing down or stopping the vehicle before a crosswalk C when it is estimated that a pedestrian A in the vicinity region R has the intention to cross the road. The control device 20 outputs control signals to actuators (brake control system, engine control system, steering control system) 40 so as to achieve the driving route generated by the route generation unit 22F.

[0026] [Probability distribution data] Next, the probability distribution data PD used to estimate pedestrian A's intention to cross will be described with reference to Figs. 3 to 6. Fig. 3 is a schematic diagram illustrating the probability distribution data PD. As will be described later, the probability distribution data PD is created based on statistical data obtained by observing the movements of many pedestrians within an area near an actual crosswalk. The probability distribution data PD sets probabilities or probability distributions at multiple positions within the area.

[0027] The probability distribution data PD can be set in a left virtual area VRa and a right virtual area VRb. The virtual area VRa is a rectangular area surrounded by four sides: a central boundary line CLa, a distal boundary line SLa extending in the y direction, and a front boundary line FLa and a rear boundary line RLa extending in the x direction (e.g., a 30 m x 30 m virtual area). A virtual gate GVa is set in the virtual area VRa, and extends from a first end point p1a on the central boundary line CLa to a second end point p2a on the front boundary line FLa, which is a predetermined width (e.g., 2 to 4 m) away along the x direction. The direction in which the virtual gate GVa extends is the x direction, and the direction perpendicular to this is the y direction. A probability or probability distribution is set at each position (x, y) within this virtual area VRa.

[0028] The virtual region VRb is equivalent to the virtual region VRa folded back to the right at the central boundary line CLa, and the two are formed symmetrically. That is, the virtual region VRb is a rectangular region surrounded by the central boundary line CLb, the anterior boundary line FLb, the distal boundary line SLb, and the posterior boundary line RLb. The virtual gate GVb extends from a first end point p1b on the central boundary line CLb to a second end point p2b on the anterior boundary line FLb, which is a predetermined width away. A probability or probability distribution is set for the virtual region VRb so that it has a line-symmetric positional relationship with the virtual region VRa.

[0029] As described above, the virtual area VRb is bilaterally symmetrical to the virtual area VRa, so that only the virtual area VRa may be stored in the storage unit 24, and the virtual area VRb may be generated from the virtual area VRa.

[0030] That is, in the probability distribution data PD of this embodiment, a gate having an original length is divided into virtual gates GVa and GVb at its center, and accordingly, a set area having an original area width is divided into virtual areas VRa and VRb at its center. This is for the following reason. Figures 4 and 5 are explanatory diagrams showing the crossing patterns or crossing trajectories of pedestrians when crossing a crosswalk.

[0031] As shown in Figure 4, if an exit gate Gc is set at the entrance to crosswalk C, connecting the left endpoint pca and right endpoint pcb in the width direction of crosswalk C, pedestrian A on the left side of crosswalk C will typically cross the area of ​​exit gate Gc near the left endpoint pca, which is closer to their current position, and will rarely cross near the right endpoint pcb, which is further away. Also, as shown in Figure 5, pedestrian A on the extension side of crosswalk C will typically proceed approximately straight across crosswalk C. Therefore, it is sufficient to set a virtual area VRa that is half the size for a virtual gate GVa that is half the length, and the other half of the area (virtual area VRb) can be set symmetrically to the set area.

[0032] Within the virtual area VRa of the probability distribution data PD, a plurality of positions p(x, y) are set in a grid pattern with the virtual gate GVa as the reference position. A predetermined probability distribution or probability is specified for each position p(x, y). The probability distribution data PD includes the velocity vector v(v x ,v y) is defined. Note that "g" is a variable representing the gate that will be passed through in the future. That is, the probability distribution Pr(v|p,g=GVa) is generated based on observation data, and defines the probability distribution of the velocity vector v of a pedestrian who will pass through the virtual gate GVa at each position p in the future. The velocity vector v indicates the pedestrian's velocity in the x and y directions. Note that the position p does not need to be set in a grid pattern, and may be a mathematical function of the positions x and y that approximates the probability distribution Pr(v|p,g=GVa). Furthermore, in this embodiment, the probability distribution data PD has the probability distribution Pr(v|p,g=GVa) of the velocity vector at each position p. In addition, it may also include the probability distribution of the acceleration vector and the probability distribution of the direction of travel.

[0033] Furthermore, the probability distribution data PD specifies, for each position p, the probability distribution Pr(v|p) of the pedestrian's velocity vector v and the probability Pr(GVa) that the pedestrian will pass through the virtual gate GVa. Here, pedestrians include pedestrians who exit the virtual area VRa without passing through the virtual gate GVa. In other words, the probability distribution Pr(v|p) is generated based on observation data, and is the probability distribution of the velocity vector v at position p of a pedestrian who exits the virtual area VRa, not just the virtual gate GVa. Furthermore, the probability Pr(GVa) (also called the "prior probability") is generated based on observation data, and is the probability that a pedestrian who has passed through position p will exit the virtual area VRa through the virtual gate GVa.

[0034] 6 is a schematic explanatory diagram in which the probability distribution data PD is applied to a crosswalk C. In the figure, the arrows indicate the average value of the velocity vector v that a pedestrian A crossing the crosswalk C can take at each position p.

[0035] [Creating probability distribution data] Next, the observation data used to create the probability distribution data PD in this embodiment will be described with reference to Figs. 7 to 10. The observation data was acquired by continuously capturing images of people flowing at the intersection shown in Fig. 7. Pedestrians were identified in the captured image data, and their movements were tracked. Fig. 8 shows the results showing the trajectories of people flowing at the intersection in Fig. 7 (each trajectory represents the movement of a pedestrian).

[0036] As shown in Figure 8, exit gates G1 to G5 are set. Each exit gate is set at a location where pedestrians are normally expected to exit the vicinity of the crosswalk. That is, exit gates G1 to G3 are set relative to the sidewalk, and exit gates G4 and G5 are set relative to the crosswalk. Therefore, the vicinity area is generally surrounded by exit gates G1 to G5. The part of the boundary of this vicinity area that is not surrounded by the exit gates is a fixed object F (such as a shrub or a wall) that obstructs pedestrian passage, or the boundary between the sidewalk and the roadway (not a crosswalk, but an area where pedestrians are prohibited from passing). Using each of these exit gates as a reference position, the position p and velocity vector v of the pedestrian were acquired as observation data. That is, the observation data includes data on pedestrians heading from the sidewalk to the crosswalk, data on pedestrians heading to the sidewalk without crossing the crosswalk, and data on pedestrians heading from the crosswalk to the sidewalk.

[0037] Based on this observation data, data on whether or not a pedestrian passed through each exit gate, and the pedestrian's position p and velocity vector v when they did pass were analyzed to create probability distribution data for each exit gate (see, for example, the left diagrams of Figures 9A and 9B). Furthermore, by taking advantage of the fact that the probability distribution is linearly symmetric with respect to the center position in the longitudinal direction of the exit gate as described above, it is possible to create probability distribution data for half the area by folding back the probability distribution data obtained at a certain gate at the center position of the gate (that is, for example, by superimposing the probability distribution data for the right half of the area with respect to the exit gate on the probability distribution data for the left half of the area), as conceptually shown in Figures 9A and 9B.

[0038] Furthermore, as explained with reference to FIG. 4, pedestrians on the side of an exit gate pass through the area near the proximal end point of the exit gate. Therefore, as shown in FIG. 9C, by preparing multiple probability distribution data for half areas such as those illustrated in FIGS. 9A and 9B, and aligning these areas so that the x and y directions and the left end points of each exit gate coincide, the respective probability distribution data can be superimposed. This creates the probability distribution data PD of this embodiment. Note that if sufficient observation data can be acquired, it is not necessary to perform the processes shown in FIGS. 9A to 9C.

[0039] An example of applying the probability distribution data PD to an actual exit gate G will be described. As shown in FIG. 10, the length dimension of the actual exit gate G may be long and may not match twice the length of the virtual gate GVa (i.e., the sum of the lengths of the virtual gates GVa and GVb). In this case, as shown in FIG. 10(a), the probability distribution data PD is applied so that the end point p2a of the virtual gate GVa in the virtual region VRa matches the end point pa on the left side as you face the exit gate G. Furthermore, the inverted probability distribution data PD is applied so that the end point p2b of the virtual gate GVb in the virtual region VRb matches the end point pb on the right side of the exit gate G.

[0040] Furthermore, based on the characteristics of pedestrians described with reference to Fig. 5, the probability distribution data of positions near the central boundaries CLa and CLb of the virtual areas VRa and VRb (probability distribution data of a narrow area extending in the y direction from the x position near the central boundary line) can be applied to the blank area between the central boundaries CLa and CLb of the virtual areas VRa and VRb. This makes it possible to apply the probability distribution data PD having a virtual area without a blank area.

[0041] 10, the actual length of the exit gate G may be shorter than twice the length of the virtual gate GVa (i.e., the sum of the lengths of the virtual gates GVa and GVb). In this case, as in the example of FIG. 10, the virtual regions VRa and VRb are aligned with the endpoints of the virtual gate and the exit gate G, and then the overlapping portion of either virtual region VRa or VRb can be deleted so that they do not overlap. This makes it possible to apply probability distribution data PD having a virtual region that is symmetrical with respect to the center position of the exit gate G and has no overlapping portion. As such, in this embodiment, there is no need to acquire observation data for crosswalks with various length dimensions, and one piece of probability distribution data PD can be applied to any exit gate, thereby reducing the storage capacity required for the memory unit 24.

[0042] [Crossing intention estimation method] Next, a method for estimating pedestrian A's crossing intention will be described with reference to FIGS. 11 to 13. FIG. 11 is an explanatory diagram of crossing intention estimation according to this embodiment. In the situation shown in FIG. 11, pedestrian A is walking in a neighborhood area R of a crosswalk C. Exit gates Ga, Gb, and Gc are set ahead of pedestrian A in the direction of travel. For example, exit gate Ga is set at the boundary between the neighborhood area R and sidewalk S in case pedestrian A continues walking on sidewalk S without crossing crosswalk C, exit gate Gb is a crosswalk gate GX set for crosswalk C, and exit gate Gc is set at the entrance to the park to leave sidewalk S and enter the park.

[0043] In this embodiment, the nearby region R can be set, for example, on the sidewalk S adjacent to the crosswalk C and within a predetermined distance (for example, 20 m) from the crosswalk C. The nearby region R can preferably be a rectangular or polygonal region along the sidewalk S extending from the crosswalk C (or a combination of multiple rectangular regions when the sidewalk S intersects with another sidewalk S).

[0044] 12A, 12B, and 12C show examples in which probability distribution data PD is applied to each of the exit gates Ga, Gb, and Gc. As can be seen from these examples, the probability distribution data PD can be aligned by rotating, inverting, and translating each of the exit gates as a reference position. Furthermore, as described with reference to FIG. 10, a long and narrow region of probability distribution may be filled between the virtual regions VRa and VRb according to the length of the exit gate (e.g., FIG. 12A), and overlapping regions may be removed (e.g., FIG. 12B). In this way, a probability field based on the probability distribution data PD is set for each of the exit gates Ga, Gb, and Gc at least in the neighborhood region R. That is, a probability field is set independently for each of the exit gates in the neighborhood region R. Therefore, at position p within the neighborhood region R, the probability fields for the exit gates Ga, Gb, and Gc are simultaneously set.

[0045] In this embodiment, the system 10 calculates, for a pedestrian A in a random field set for the exit gate Ga, the likelihood La (= Pr(v|p, g=Ga)) of passing through the exit gate Ga at each position p in the neighborhood region R based on the probability distribution data PD using Bayesian estimation. Similarly, for the exit gates Gb and Gc, the system 10 calculates the likelihoods Lb and Lc (= Pr(v|p, g=Gb)), Lc (= Pr(v|p, g=Gc)) of passing through the exit gates Gb and Gc at each position p in the neighborhood region R. The system 10 estimates that the pedestrian will pass through the exit gate with the greatest likelihood among these likelihoods.

[0046] In the example of FIG. 11, assume that at pedestrian A's current location, likelihood La is 0.1, likelihood Lb is 0.3, and likelihood Lc is 0.01. In this case, since likelihood Lb is the maximum likelihood, system 10 estimates that pedestrian A will pass through exit gate Gb (i.e., cross crosswalk C). Therefore, if the exit gate of crosswalk C has the maximum likelihood among multiple exit gates set in the vicinity region R of crosswalk C (including exit gates set for the crosswalk), system 10 estimates that pedestrian A intends to cross crosswalk C. Note that an additional requirement for estimating an intention to cross may be that the maximum likelihood is equal to or greater than a predetermined likelihood threshold LT.

[0047] Furthermore, when the exit gate of the crosswalk C has the maximum likelihood, the ratio of the maximum likelihood to the sum of the likelihoods Lk of all the exit gates Gk is estimated as the probability that pedestrian A will cross the crosswalk C. In this case, an additional requirement for estimating an intention to cross may be that this probability is equal to or greater than a predetermined probability threshold.

[0048] FIG. 13 shows the results of estimating which of two exit gates G10 and G11 a large number of pedestrians will pass through, using the system 10 of this embodiment. FIG. 13 shows the trajectories of the pedestrians. The pedestrians moved from the bottom to the top in FIG. 13 and passed through either exit gate G10 or G11. Most of the observed pedestrians were walking within a normal movement speed range. In the example of FIG. 13, the pedestrians were not estimated at the position indicated by the dashed-dotted line, but they were estimated to pass through exit gate G10 at the position indicated by the solid line, and they were estimated to pass through exit gate G11 at the position indicated by the dotted line. In the example of FIG. 13, the accuracy rate of correct estimation 5 seconds before passing through the exit gate was approximately 85%.

[0049] [Process flow for estimating crossing intention] Next, the processing flow of the crossing intention estimation process of this embodiment will be described with reference to Figures 14 and 15. In this embodiment, the control device 20 is configured to execute the crossing intention estimation process while the vehicle 1 is in autonomous driving. When the vehicle 1 starts, the control device 20 repeatedly executes the processing flows of Figures 14 and 15. When the autonomous driving control is initiated after the vehicle 1 starts, the control device 20 continuously executes the autonomous driving process by constantly calculating the vehicle position using the host vehicle position estimation unit 22A and controlling the actuator 40 so that the vehicle 1 travels along the travel route generated by the route generation unit 22F (S11).

[0050] Next, the control device 20 determines whether or not a crosswalk C without a traffic light has been detected by the crosswalk detection unit 22B while the vehicle is traveling (S12). If the determination is negative (S12; NO), the process ends. On the other hand, if the determination is positive (S12; YES), the control device 20 determines whether or not the pedestrian detection unit 22D has detected the presence of a pedestrian A on the crosswalk C (S13). If the determination is positive (S13; YES), the control device 20 updates the travel route using the route generation unit 22F to stop the vehicle just before the crosswalk C, and controls the actuator 40 to decelerate the vehicle toward the crosswalk C according to this travel route (S17), and then ends the process.

[0051] On the other hand, if the determination is negative (S13; NO), the control device 20 determines whether or not the pedestrian detection unit 22D has detected the presence of pedestrian A in the nearby region R on the sidewalk S near the crosswalk C (S14). If the determination is negative (S14; NO), the process ends. On the other hand, if the determination is positive (S14; YES), the control device 20 determines, using the crossing-intention estimation unit 22E, whether or not pedestrian A intends to cross the crosswalk C (S15).

[0052] If it is determined that the vehicle has an intention to cross (S16; YES), the control device 20 controls the actuator 40 to slow down or stop the vehicle in front of the crosswalk C according to the new travel route updated by the route generation unit 22F (S17), and ends the process. On the other hand, if it is determined that the vehicle has no intention to cross (S16; NO), the control device 20 ends the process without slowing down or stopping the vehicle.

[0053] In step S17, instead of automatically braking the vehicle 1, the control device 20 may notify the driver by the notification device 50 that the pedestrian A is about to cross the crosswalk C and that the vehicle 1 should therefore be slowed down or stopped.

[0054] Next, the crossing intention estimation process in step S16 will be described with reference to Fig. 15. First, the control device 20 sets multiple exit gates Gk (k = 1, 2, ... N) in the vicinity region R set for the crosswalk C (multiple crosswalks are possible) using the gate setting unit 22C (S21). Next, the control device 20 executes the following process (steps S22 to S24) for each exit gate Gk using the crossing intention estimation unit 22E.

[0055] The control device 20 sets relative coordinates for the exit gate Gk based on the positional relationship between the exit gate Gk and the pedestrian A, in addition to the absolute coordinates of the nearby region R (S22). That is, since each exit gate Gk has a different orientation, a unique relative coordinate is set for each exit gate Gk.

[0056] Next, the control device 20 applies the probability distribution data PD to the exit gate Gk based on the set relative coordinates (S23). That is, as described above, the virtual regions VRa and VRb are applied to the nearby region R. At this time, the direction in which the exit gate Gk and the virtual gate GVa of the virtual region VRa extend is made to coincide with each other, and the left end point Pa of the exit gate Gk is made to coincide with the second end point p2a of the virtual gate GVa. Furthermore, the direction in which the exit gate Gk and the virtual gate GVb of the virtual region VRb extend is made to coincide with each other, and the right end point pb of the exit gate Gk is made to coincide with the second end point p2b of the virtual gate GVb of the virtual region VRb.

[0057] Then, the control device 20 generates probability distribution data PD for the exit gate Gk based on the aligned virtual areas VRa and VRb. At this time, as described with reference to Fig. 10, if there is a blank area between the virtual areas VRa and VRb, the probability distribution data near the central boundary lines of the virtual areas VRa and VRb is applied to this blank area, and if the virtual areas VRa and VRb overlap, one of the overlapping areas is deleted.

[0058] Next, the control device 20 calculates the likelihood Lk that the detected pedestrian A will pass through the exit gate Gk based on the probability distribution data PD set for the exit gate Gk (S24). The control device 20 determines whether the likelihood Lk has been calculated for all exit gates Gk (S25). If the determination is affirmative (S25; YES), the process proceeds to step S26, and if the determination is negative (S25; NO), the process returns to step S22 and executes the processes of steps S22 to S24 for the next exit gate Gk.

[0059] Next, the control device 20 selects the exit gate Gk with the largest likelihood Lk from the calculated likelihoods Lk (k = 1 to N) using the crossing-intention estimation unit 22E (S26), and determines whether the selected exit gate Gk is set for the crosswalk C (S27). If the determination is negative (S27; NO), the control device 20 uses the crossing-intention estimation unit 22E to estimate that the pedestrian A does not intend to cross (the pedestrian A will proceed to a location other than the crosswalk C) (S30), and ends the process (return to step S16).

[0060] On the other hand, if the determination is affirmative (S27; YES), the control device 20 determines whether the likelihood Lk of the selected exit gate Gk is equal to or greater than a predetermined likelihood threshold LT using the crossing-intention estimation unit 22E (S28). If the determination is negative (S28; NO), it is determined that the pedestrian A does not intend to cross (S28), and the process ends. On the other hand, if the determination is affirmative (S28; YES), the control device 20 determines, using the crossing-intention estimation unit 22E, that the pedestrian A intends to cross (S29), and the process ends.

[0061] Next, the effects of the vehicle crossing intention estimation system 10 according to this embodiment will be described. The vehicle crossing intention estimation system 10 according to this embodiment includes a crosswalk detection unit 22B that detects a crosswalk C ahead of the vehicle 1 on a travel lane L; a gate setting unit 22C that virtually sets multiple exit gates Gk for exiting from the virtual area R to the virtual area R around the detected crosswalk C, which includes a sidewalk S connected to the detected crosswalk C; a memory unit 24 that stores probability distribution data PD including at least a probability distribution Pr(v|p) of velocity vectors v within the virtual areas VRa and VRb of pedestrians who exit the virtual areas VRa and VRb from the virtual areas VRa and VRb through the virtual gates GVa and GVb; a pedestrian detection unit 22D that detects a pedestrian A within the virtual area R around the crosswalk C; and a crossing intention estimation unit 22E that estimates whether the pedestrian A detected within the virtual area R intends to cross the crosswalk C. The gate setting unit 22C sets multiple exit gates Gk for exiting the virtual area R from the virtual area VRa and VRb to the virtual area R. A crosswalk gate GX is set at the crosswalk C as one of the exit gates Gk, and the probability distribution data PD has, for each of multiple positions p within the virtual areas VRa and VRb relative to the virtual gates GVa and GVb, a probability distribution Pr(v|p) of the velocity vector v of a pedestrian exiting the virtual areas VRa and VRb from the virtual gates GVa and GVb, as well as a probability distribution Pr(v) of the pedestrian velocity vector v and a probability Pr(GV) of the pedestrian passing through the virtual gates GVa and GVb. The crossing intention estimation unit 22E independently applies the probability distribution data PD to each of the multiple exit gates Gk to calculate the likelihood Lk that a pedestrian A walking within the nearby area R will pass through each exit gate Gk, and if the exit gate Gk with the highest likelihood Lk is the crosswalk gate GX, it estimates that pedestrian A within the nearby area R has the intention to cross the crosswalk C.

[0062] According to this embodiment configured as described above, it is possible to estimate whether pedestrian A intends to cross the crosswalk C by referring to the probability distribution data of the velocity vector v of pedestrian A near the crosswalk C. Therefore, this embodiment can estimate the pedestrian's intention to cross the crosswalk C with higher accuracy than conventional methods that estimate the pedestrian's intention to cross based only on limited information such as the pedestrian's position and behavior. Furthermore, this embodiment can avoid unnecessarily slowing down or stopping vehicles at crosswalks without traffic lights.

[0063] Furthermore, in this embodiment, preferably, the gate setting unit 22C does not set an exit gate Gk at a ground fixed object F on the boundary of the nearby area R that would prevent the pedestrian A from leaving the nearby area R, but sets at least a crosswalk gate GX (Gk) for crossing the crosswalk C, and an exit gate Gk at the boundary between the sidewalk S and the nearby area R. According to this embodiment configured in this way, the pedestrian A can go from the nearby area R to the outside of the nearby area R through any one of the exit gates Gk.

[0064] In this embodiment, the probability distribution data PD is preferably statistical data based on observation data of pedestrians walking on a predetermined sidewalk S that connects to a predetermined crosswalk C serving as virtual gates GVa and GVb. According to this embodiment configured in this manner, it is possible to set a probability distribution of the speed vector v of pedestrian A that selects the crosswalk C for the nearby region R based on actual observation data.

[0065] Furthermore, in this embodiment, preferably, when the crossing intention estimation unit 22E estimates that pedestrian A in the vicinity region R has an intention to cross (S16; YES), it activates the braking device (actuator 40) of the vehicle 1 or issues a predetermined notification to the driver of the vehicle 1. According to this embodiment configured as above, when it is estimated that pedestrian A has an intention to cross at a crosswalk C without a traffic light, it is possible to decelerate the vehicle 1. As a result, in this embodiment, it is possible to avoid unnecessarily decelerating the vehicle 1 when it is not estimated that pedestrian A has an intention to cross.

[0066] In this embodiment, the virtual region VRa of the probability distribution data PD preferably has a center boundary line CLa and extends from the center boundary line CLa in a predetermined direction (-x direction) perpendicular to the center boundary line CLa. The virtual gate GVa extends from a first endpoint p1a on the center boundary line CLa to a second endpoint p2a, a predetermined width away from the center boundary line CLa, in the predetermined direction. The crossing intention estimation unit 22E forms another virtual region VRb with a probability distribution that is line-symmetrical to the virtual region VRa with respect to the center boundary line CLa, and applies the probability distribution data PD of the virtual region VRa and the other virtual region VRb to the exit gate Gk. According to this embodiment, the memory unit 24 stores the probability distribution data of only one virtual region (e.g., the left side of the virtual gate) in advance. When applying the probability distribution data to the exit gate Gk, a separate virtual region with an inverted probability distribution is formed. Therefore, this embodiment can reduce the volume of the probability distribution data PD stored in the memory unit 24.

[0067] In this embodiment, the crossing-intention estimation unit 22E preferably aligns the second end points p2a and p2b of the virtual gates GVa and GVb of the virtual area VRa and another virtual area VRb so that they coincide with the two end points pa and pb of the exit gate Gk, and applies the probability distribution data PD to the exit gate Gk. According to this embodiment configured as described above, the pre-stored probability distribution data PD can be commonly used for exit gates Gk that face in various directions and have various length dimensions. [Explanation of symbols]

[0068] 1 vehicle 10 Vehicle Crossing Intention Estimation System 20 Control device A. Pedestrians C. Crosswalk GVa,GVb virtual gates PD Probability Distribution Data R neighborhood S Sidewalk VRa,VRb virtual region

Claims

1. A vehicle crossing estimation system, comprising: a crosswalk detection unit that detects a crosswalk without a traffic light ahead of a vehicle on a travel lane; a gate setting unit that virtually sets a plurality of exit gates for exiting from within a vicinity area of ​​the detected crosswalk, the vicinity area including a sidewalk connected to the crosswalk; and a storage unit that stores probability distribution data including at least a probability distribution of a velocity vector of a pedestrian who leaves a virtual area through a virtual gate and exits the virtual area; a pedestrian detection unit that detects pedestrians within the vicinity of the crosswalk; a crossing intention estimation unit that estimates whether a pedestrian detected within the vicinity area has an intention to cross the crosswalk, the gate setting unit sets a crosswalk gate for the crosswalk as one of the plurality of exit gates, the probability distribution data includes, for each of a plurality of positions in the virtual area relative to the virtual gate, a probability distribution of the velocity vector of a pedestrian exiting the virtual area from the virtual gate, and a probability distribution of a velocity vector of the pedestrian and a probability that the pedestrian will pass through the virtual gate; the crossing-intention estimation unit applies the probability distribution data independently to each of the plurality of exit gates to calculate the likelihood that a pedestrian walking within the vicinity area will pass through each exit gate, and if the exit gate with the greatest likelihood is the pedestrian crossing gate, estimates that the pedestrian within the vicinity area has the intention to cross the pedestrian crossing.

2. 2. The vehicle crossing estimation system according to claim 1, wherein the gate setting unit does not set the exit gate at a fixed object on the ground at a boundary of the vicinity area that prevents pedestrians from exiting the vicinity area, but sets an exit gate at least at the crosswalk gate for crossing the crosswalk and at the boundary between the sidewalk and the vicinity area.

3. 2. The vehicle crossing estimation system according to claim 1, wherein the probability distribution data is statistical data based on observation data of pedestrians walking on a predetermined sidewalk connected to the predetermined crosswalk serving as the virtual gate.

4. 2. The vehicle crossing estimation system according to claim 1, wherein the crossing intention estimation unit, when estimating a pedestrian's intention to cross within the vicinity area, activates a braking device of the vehicle or issues a predetermined notification to a driver of the vehicle.

5. the virtual region of the probability distribution data has a central boundary line and extends from the central boundary line in a predetermined direction perpendicular to the central boundary line, and the virtual gate extends from a first end point on the central boundary line to a second end point spaced a predetermined width apart in the predetermined direction; 2. The vehicle crossing intention estimation system according to claim 1, wherein the crossing intention estimation unit forms another virtual area having a probability distribution that is line-symmetrical to the virtual area with respect to the center boundary line, and applies the probability distribution data of the virtual area and the another virtual area to the exit gate.

6. 6. The vehicle crossing intention estimation system according to claim 5, wherein the crossing intention estimation unit aligns the second end points of the virtual gates in the virtual area and the another virtual area to coincide with two end points of the exit gate, and applies the probability distribution data to the exit gate.

Images