Pedestrian crossing prediction method and pedestrian crossing prediction device

Abstract

To provide a pedestrian traverse prediction method and pedestrian traverse prediction device capable of accurately predicting whether pedestrians are to cross a road when the pedestrians exist on both sides of the road.SOLUTION: A pedestrian traverse prediction device includes an object detection device and controller mounted on an own vehicle 50. The object detection device detects a pedestrian 60 within a predetermined distance in front of the own vehicle 50. The object detection device detects a pedestrian 61 on the opposite side of a road with a position of the pedestrian 60 as reference. When a behavior change of the pedestrian 61 is detected within a predetermined time after the object detection device detects a behavior change of the pedestrian 60, the controller allows at least one likelihood between a first likelihood when the pedestrian 60 crosses the road and a second likelihood when the pedestrian 61 crosses the road to be higher than that when the behavior change of the pedestrian 61 is not detected within the predetermined time.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a pedestrian crossing prediction method and a pedestrian crossing prediction device. [Background technology]

[0002] Conventionally, when turning right or left at an intersection, a method is known for predicting the possibility that a vehicle and a pedestrian will cross paths based on the traveling direction and speed of the vehicle after turning right or left and the traveling direction and speed of the pedestrian (Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2014-232412 Summary of the Invention [Problem to be solved by the invention]

[0004] However, the invention described in Patent Document 1 does not take into account the possibility that when pedestrians are present on both sides of the road, one pedestrian may accelerate to cross the crosswalk in response to a change in the behavior of the other pedestrian, which could result in a decrease in prediction accuracy.

[0005] The present invention has been made in consideration of the above-mentioned problems, and its purpose is to provide a pedestrian crossing prediction method and a pedestrian crossing prediction device that can accurately predict whether a pedestrian will cross a road when pedestrians are present on both sides of the road. [Means for solving the problem]

[0006] A pedestrian crossing prediction method according to one embodiment of the present invention detects a first pedestrian within a predetermined distance ahead of a vehicle, and detects a second pedestrian on the other side of the road based on the position of the first pedestrian.If a change in the behavior of the second pedestrian is detected within a predetermined time after a change in the behavior of the first pedestrian is detected, at least one of the first likelihood that the first pedestrian will cross the road and the second likelihood that the second pedestrian will cross the road is increased compared to a case where a change in the behavior of the second pedestrian is not detected within the predetermined time. [Effects of the Invention]

[0007] According to the present invention, it is possible to accurately predict whether a pedestrian will cross a road when there are pedestrians on both sides of the road. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a configuration diagram of a pedestrian crossing prediction device according to an embodiment of the present invention. [Figure 2] FIG. 2 is a diagram illustrating an example of a method for calculating likelihood. [Figure 3] FIG. 3 is a diagram illustrating an example of a method for calculating likelihood. [Figure 4] FIG. 4 is a diagram illustrating an example of a method for calculating the likelihood. [Figure 5] FIG. 5 is a flowchart illustrating an example of the operation of the pedestrian crossing prediction device. [Figure 6] FIG. 6 is a diagram illustrating an example of a method for calculating the likelihood. [Figure 7] FIG. 7 is a diagram illustrating an example of a method for calculating likelihood. [Figure 8] FIG. 8 is a diagram illustrating an example of a method for calculating the likelihood. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the description of the drawings, the same parts are designated by the same reference numerals and the description thereof will be omitted.

[0010] An example of the configuration of a pedestrian crossing prediction device will be described with reference to Fig. 1. As shown in Fig. 1, the pedestrian crossing prediction device includes an object detection device 1, a vehicle position estimation device 2, a map database 3, and a controller 100.

[0011] The pedestrian crossing prediction device is installed in a vehicle (host vehicle). The pedestrian crossing prediction device may be installed in a vehicle with an automatic driving function, or in a vehicle without an automatic driving function. The pedestrian crossing prediction device may also be installed in a vehicle that is capable of switching between automatic driving and manual driving. The automatic driving function may also be a driving assistance function that automatically controls only some of the vehicle control functions, such as steering control, braking force control, and driving force control, to assist the driver in driving.

[0012] The object detection device 1 is composed of object detection sensors such as a laser range finder, radar, lidar, camera, and sonar. The object detection device 1 detects objects around the host vehicle using multiple object detection sensors. The object detection device 1 detects moving objects including other vehicles, buses, motorbikes, bicycles, and pedestrians, as well as stationary objects including crosswalks and bus stops. The object detection device 1 detects the position, attitude, size, speed, acceleration, etc. of moving and stationary objects relative to the host vehicle. The object detection device 1 outputs information about the detected objects to the controller 100. Note that object detection methods using laser range finders, radar, lidar, cameras, sonar, etc. are well known, so a detailed description will be omitted.

[0013] The vehicle position estimation device 2 detects the position (position information) of the vehicle on the ground. One example of the vehicle position estimation device 2 is a GPS receiver. The GPS receiver detects the vehicle's position information on the ground by receiving radio waves from artificial satellites. The vehicle's position information detected by the GPS receiver includes latitude information and longitude information. The GPS receiver outputs the detected vehicle's position information to the controller 100. Note that the method of detecting the vehicle's position information is not limited to a GPS receiver. For example, the vehicle's position may be estimated using a method called odometry. Odometry is a method of estimating the vehicle's position by determining the amount and direction of movement of the vehicle according to the rotation angle and rotation angular velocity of the vehicle. Note that a GNSS receiver may be used instead of the GPS receiver.

[0014] The controller 100 is a general-purpose microcomputer equipped with a CPU (Central Processing Unit), memory, and input / output units. A computer program for functioning as a pedestrian crossing prediction device is installed in the microcomputer. By executing the computer program, the microcomputer functions as multiple information processing circuits included in the pedestrian crossing prediction device. Note that while an example of implementing the multiple information processing circuits included in the pedestrian crossing prediction device using software is shown here, it is of course possible to configure the information processing circuits by providing dedicated hardware for executing each of the information processes described below. Alternatively, the multiple information processing circuits may be configured as separate hardware. The controller 100 includes, as examples of multiple information processing circuits, a detection integration unit 4, an object tracking unit 5, an in-map position estimation unit 6, a behavior prediction unit 10, and a vehicle control unit 20. The behavior prediction unit 10 is further divided into a pedestrian detection unit 11, a behavior change detection unit 12, a gaze detection unit 13, an attribute detection unit 14, a statistical data acquisition unit 15, and a likelihood calculation unit 16.

[0015] The detection integration unit 4 integrates the multiple detection results obtained from each of the multiple object detection sensors included in the object detection device 1 and outputs a single detection result for each object. Specifically, it calculates the most reasonable object behavior that minimizes error from the object behavior obtained from each object detection sensor, taking into account the error characteristics of each object detection sensor. Specifically, by using well-known sensor fusion technology, it comprehensively evaluates the detection results obtained from multiple types of sensors to obtain more accurate detection results.

[0016] The object tracking unit 5 tracks the objects detected by the detection integration unit 4. Specifically, the object tracking unit 5 verifies (associates) the identity of the objects between different times based on the behavior of the objects output at different times, and tracks the objects based on the association.

[0017] The intra-map position estimation unit 6 estimates the position of the host vehicle on a map based on the absolute position of the host vehicle obtained by the host vehicle position estimation device 2 and the map information acquired from the map database 3. The map database 3 is a database stored in a car navigation device or the like, and stores map information necessary for route guidance, such as road information and facility information. The road information includes, for example, information about intersections, the number of lanes on roads, road boundaries, and lane connections. The map database 3 outputs map information to the intra-map position estimation unit 6 in response to a request from the intra-map position estimation unit 6. In this embodiment, the pedestrian crossing prediction device is described as having the map database 3, but the pedestrian crossing prediction device does not necessarily have to have the map database 3. The map information may be acquired by various sensors such as a camera, or may be acquired using vehicle-to-vehicle communication or road-to-vehicle communication. Furthermore, if the map information is stored in an external server, the pedestrian crossing prediction device may acquire the map information from the server via communication as needed. Furthermore, the pedestrian crossing prediction device may periodically obtain the latest map information from the server and update the map information it holds. The map information stored in the map database 3 may be high-definition map data (HD MAP) or normal map data (SD MAP).

[0018] Next, the function of the behavior prediction unit 10 will be described in detail with reference to Figures 2 to 7. The scene shown in Figure 2 is a scene in which the host vehicle 50 is traveling toward an intersection. Reference numerals 60 and 61 indicate pedestrians.

[0019] As shown in FIG. 2, when a pedestrian 60 is present on one side of a crosswalk and a pedestrian 61 is present on the other side, the pedestrians 60 and 61 are detected by the pedestrian detection unit 11 (step S101 in FIG. 5). More specifically, the pedestrian detection unit 11 detects the pedestrians 60 and 61 using data acquired from the object tracking unit 5. When the pedestrians 60 and 61 are detected on both sides of the crosswalk (YES in steps S101 and S103 in FIG. 5), the pedestrian detection unit 11 outputs information related to the detection to the behavior change detection unit 12. In FIG. 2, it is assumed that the pedestrian 60 is detected on the left side of the crosswalk as viewed from the front of the drawing, and the pedestrian 61 is detected on the right side of the crosswalk. Note that the pedestrian 61 may also be detected on the other side of the crosswalk in the direction in which the pedestrian 60 is traveling (to the right in FIG. 2). Furthermore, since a detection range is set for the sensors included in the object detection device 1, it is assumed that the pedestrians 60 and 61 are detected within a predetermined distance ahead of the host vehicle 50.

[0020] The behavior change detection unit 12 detects behavior changes of the pedestrians 60 and 61 detected by the pedestrian detection unit 11. In this embodiment, "pedestrian behavior" includes the pedestrian's speed, gestures, hand gestures, facial orientation, and body orientation. Therefore, "pedestrian behavior change" refers to a change in the pedestrian's walking speed, gestures, hand gestures, facial orientation, or body orientation. In the example of FIG. 2, assume that a behavior change in which the pedestrian 60 waves his / her hand is first detected (step S105 in FIG. 5). After the hand gesture of the pedestrian 60 is detected, the behavior change detection unit 12 determines whether a behavior change of the pedestrian 61 is detected within a predetermined time. The "predetermined time" here is, for example, two seconds. Here, assume that a hand gesture (behavior change) of the pedestrian 61 is detected within the predetermined time in the same manner as the hand gesture of the pedestrian 60 (YES in step S107 in FIG. 5). In this case, the likelihood that the pedestrian 61 will cross the crosswalk is calculated to be higher than when the hand gesture of the pedestrian 61 is not detected. This is because, if a hand gesture from pedestrian 61 is detected within a predetermined time after a hand gesture from pedestrian 60, it is highly likely that pedestrian 61 and pedestrian 60 are friends or acquaintances, and it is estimated that pedestrian 61 is likely to cross the crosswalk to meet pedestrian 60. In this embodiment, "likelihood" refers to plausibility and has a concept of high and low. For example, "the likelihood that pedestrian 61 will cross the crosswalk is high" means "the likelihood that pedestrian 61 will cross the crosswalk." The likelihood is calculated by the likelihood calculation unit 16. Note that "the likelihood is calculated to be high" and "the likelihood is calculated to be low" mean, for example, compared to the case where no causal event exists. The behavior change detection unit 12 outputs information related to the detection to the gaze detection unit 13. Note that, while the following mainly describes an example of calculating the "likelihood that pedestrian 61 will cross the crosswalk," the "likelihood that pedestrian 60 will cross the crosswalk" may also be calculated using a similar method. In this embodiment, it is sufficient to calculate at least one of the likelihood that the pedestrian 60 will cross the crosswalk and the likelihood that the pedestrian 61 will cross the crosswalk. Of course, both likelihoods may be calculated.

[0021] When a behavior change of one pedestrian is detected within a predetermined time after a behavior change of the other pedestrian is detected, the gaze detection unit 13 detects the gazes of both pedestrians around the timing when the behavior change of the other pedestrian occurred. Explaining with reference to FIG. 2, the gaze detection unit 13 detects the gazes of pedestrians 60 and 61 because a hand gesture of pedestrian 61 is detected within a predetermined time after a hand gesture of pedestrian 60 is detected (step S109 in FIG. 5). The gaze detection unit 13 detects the gazes of pedestrians 60 and 61 using information detected by the object detection device 1. A method of detecting gazes using a camera image or the like is known, and a detailed description thereof will be omitted. Alternatively, the gaze detection may be estimated based on the facial orientation and body orientation of pedestrians 60 and 61.

[0022] The gaze detection unit 13 determines whether the gazes of the pedestrians 60 and 61 met based on the detection results of the gazes of the pedestrians 60 and 61 (step S111 in FIG. 5). The likelihood calculation unit 16 changes the likelihood that the pedestrian 61 will cross the crosswalk depending on whether the gazes of the pedestrians 60 and 61 met. If it is determined that the gazes of the pedestrians 60 and 61 did not meet, the likelihood calculation unit 16 lowers the increased likelihood or returns it to its original state. If "the gazes of the pedestrians 60 and 61 did not meet," it is estimated that the hand gestures of the pedestrian 61 were unrelated to the hand gestures of the pedestrian 60. As a result, it is estimated that the pedestrians 60 and 61 are not friends or acquaintances, and therefore it can be said that the possibility that the pedestrian 61 will cross the crosswalk to meet the pedestrian 60 is low. Therefore, if it is determined that the gazes of the pedestrians 60 and 61 did not meet, the likelihood calculation unit 16 lowers the increased likelihood or returns it to its original state.

[0023] On the other hand, when the gaze detection unit 13 determines that the gazes of the pedestrians 60 and 61 have met, the likelihood calculation unit 16 changes the likelihood that the pedestrian 61 will cross the crosswalk based on a change in the speed of the pedestrian 61. For example, when it is detected that the pedestrian 61 has accelerated, the likelihood calculation unit 16 may increase the likelihood that the pedestrian 61 will cross the crosswalk compared to before the acceleration. Furthermore, when it is detected that the pedestrian 61 has decelerated, the likelihood calculation unit 16 may decrease the likelihood that the pedestrian 61 will cross the crosswalk compared to before the deceleration. Note that the likelihood calculation unit 16 may increase the likelihood of one of the pedestrians 60 and 61 for which acceleration related to walking has been detected compared to the likelihood of the other for which acceleration has not been detected. Furthermore, the likelihood calculation unit 16 may increase the likelihood of one of the pedestrians 60 and 61 for which deceleration related to walking has been detected compared to the likelihood of the other for which deceleration has not been detected.

[0024] If the gaze detection unit 13 determines that the gazes of the pedestrians 60 and 61 have met (YES in step S111 in FIG. 5), the process proceeds to step S113 in FIG. 5. The attribute detection unit 14 acquires the attributes of the pedestrians 60 and 61. In this embodiment, the "pedestrian attributes" include physical characteristics such as height, body width, adult, child, and elderly. The attribute detection unit 14 acquires the attributes of the pedestrians 60 and 61 using information detected by the object detection device 1. If the attribute of pedestrian 60 is detected to be "adult" and the attribute of pedestrian 61 is detected to be "child," the likelihood that pedestrian 61 will cross the crosswalk may be higher than the likelihood that pedestrian 60 will cross the crosswalk.

[0025] Once the attribute acquisition is complete, the process proceeds to step S115 in FIG. 5. The statistical data acquisition unit 15 acquires statistical data. In this embodiment, "statistical data" refers to data indicating the direction in which the majority of people cross the crosswalk depending on the time of day. For example, consider a case where a station 70 is located at the location shown in FIG. 3. In this case, in the morning, many pedestrians cross the crosswalk to head toward the station 70. That is, as shown in FIG. 3, many pedestrians cross the crosswalk from left to right. On the other hand, in the evening, contrary to the morning, many pedestrians cross the crosswalk from right to left to head away from the station 70, as shown in FIG. 4. In this manner, "statistical data" in this embodiment refers to statistical data regarding the direction in which pedestrians travel depending on the time of day. There are no particular limitations on the method of acquiring statistical data. However, if the statistical data is pre-stored in a storage device (not shown), the statistical data acquisition unit 15 can acquire the statistical data by referencing the storage device. Alternatively, the statistical data may be acquired via road-to-vehicle communication. The likelihood calculation unit 16 may calculate the likelihood using the statistical data. For example, in the morning, the likelihood calculation unit 16 may set the likelihood that pedestrian 60 will cross the crosswalk higher than the likelihood that pedestrian 61 will cross the crosswalk. On the other hand, in the evening, the likelihood calculation unit 16 may set the likelihood that pedestrian 61 will cross the crosswalk higher than the likelihood that pedestrian 60 will cross the crosswalk.

[0026] The likelihood calculation unit 16 calculates the likelihood that the pedestrians 60, 61 will cross the crosswalk based on behavior changes, line of sight, walking speed, pedestrian attributes, statistical data, etc. (step S117 in FIG. 5). As described above, the likelihood calculation unit 16 changes the likelihood that the pedestrians 60, 61 will cross the crosswalk based on each detected element, and adds up the likelihoods calculated based on each element. If the total likelihood value exceeds a predetermined value (0.7, for example), the likelihood calculation unit 16 predicts that the target pedestrian of the pedestrians 60, 61 will cross the crosswalk. The likelihood calculation unit 16 outputs the prediction result to the vehicle control unit 20.

[0027] When the likelihood calculation unit 16 predicts that one of the pedestrians 60 and 61 will cross the crosswalk, the vehicle control unit 20 accelerates the deceleration of the host vehicle 50 and stops the host vehicle 50 in front of the crosswalk (step S119 in FIG. 5). As a result, for example, when the host vehicle 50 is a vehicle equipped with an automatic driving function, the sense of discomfort felt by the occupants regarding the automatic driving is eliminated or reduced.

[0028] (Action and effect) As described above, the pedestrian crossing prediction device according to this embodiment provides the following advantageous effects.

[0029] The pedestrian crossing prediction device includes a sensor (object detection device 1) mounted on the host vehicle 50 and a controller 100. The sensor detects a pedestrian 60 (first pedestrian) within a predetermined distance ahead of the host vehicle 50. The sensor detects a pedestrian 61 (second pedestrian) on the other side of the road based on the position of the pedestrian 60. If a behavior change of the pedestrian 61 is detected within a predetermined time after the sensor detects a behavior change of the pedestrian 60, the controller 100 increases at least one of the first likelihood that the pedestrian 60 will cross the road and the second likelihood that the pedestrian 61 will cross the road, compared to when a behavior change of the pedestrian 61 is not detected within the predetermined time. This makes it possible to accurately predict whether at least one of pedestrians 60 and 61 will cross the crosswalk when pedestrians 60 and 61 are present on both sides of the crosswalk, as shown in FIG. 2 .

[0030] When a behavior change of pedestrian 61 (second pedestrian) is detected within a predetermined time after a behavior change of pedestrian 60 (first pedestrian) is detected by a sensor, controller 100 determines whether or not the gazes of pedestrian 60 and pedestrian 61 met before and after the timing when the behavior change of pedestrian 61 occurred. Controller 100 changes at least one of the first likelihood and the second likelihood based on the determination result of whether or not the gazes met. By determining whether or not the gazes met at such timing, it becomes possible to estimate whether pedestrian 60 and pedestrian 61 are friends or acquaintances. Therefore, by changing the likelihood using the determination result of whether or not the gazes met, it becomes possible to accurately predict whether at least one of the pedestrians will cross the crosswalk.

[0031] If it is determined that the eyes of pedestrians 60 and 61 did not meet, the controller 100 may lower the likelihood that it was increased. If the eyes of pedestrians 60 and 61 did not meet, it is estimated that the hand gesture of pedestrian 61 was unrelated to the hand gesture of pedestrian 60. As a result, it is estimated that pedestrians 60 and 61 are not friends or acquaintances, and it can be said that the possibility that pedestrian 61 will cross the crosswalk to meet pedestrian 60 is low. Therefore, if it is determined that the eyes of pedestrians 60 and 61 did not meet, the controller 100 may lower the likelihood that it was increased. This makes it possible to accurately predict whether at least one of the pedestrians will cross the crosswalk.

[0032] When it is determined that the gazes of the pedestrians 60 (first pedestrian) and 61 (second pedestrian) are met, the controller 100 may change at least one of the first likelihood and the second likelihood based on a change in walking speed of the pedestrians 60 (first pedestrian) and 61 (second pedestrian). By changing the likelihood based on a change in walking speed, it becomes possible to accurately predict whether at least one of the pedestrians will cross the crosswalk.

[0033] The controller 100 may increase the likelihood of one of the pedestrians 60 (first pedestrian) and 61 (second pedestrian) for which walking-related acceleration has been detected, compared to the likelihood of the other for which acceleration has not been detected. For example, if acceleration of pedestrian 61 is detected but acceleration of pedestrian 60 is not detected, the controller 100 may increase the likelihood that pedestrian 61 will cross the crosswalk more than the likelihood that pedestrian 60 will cross the crosswalk. Detection of walking-related acceleration presumes a high probability that pedestrian 61 will cross the crosswalk to meet pedestrian 60. Therefore, by increasing the likelihood depending on the acceleration, it becomes possible to accurately predict whether at least one of the pedestrians will cross the crosswalk.

[0034] The controller 100 may lower the likelihood of one of the pedestrians 60 (first pedestrian) and 61 (second pedestrian) for which deceleration related to walking has been detected, compared to the likelihood of the other for which deceleration has not been detected. For example, if deceleration of pedestrian 61 is detected but deceleration of pedestrian 60 is not detected, the controller 100 may lower the likelihood of pedestrian 61 crossing the crosswalk compared to the likelihood of pedestrian 60 crossing the crosswalk. Detection of deceleration related to walking presumes that there is a high possibility that pedestrian 61 has given up on crossing the crosswalk. Therefore, by lowering the likelihood according to deceleration, it becomes possible to accurately predict whether at least one of the pedestrians will cross the crosswalk.

[0035] The controller 100 may increase the likelihood of one of the pedestrians 60 (first pedestrian) and 61 (second pedestrian), whose attribute is detected as a child, more than the likelihood of the other, whose attribute is not detected as a child. For example, if the attribute of pedestrian 61 is detected as a child, while the attribute of pedestrian 60 is detected as an adult, the controller 100 may increase the likelihood that pedestrian 61 will cross the crosswalk more than the likelihood that pedestrian 60 will cross the crosswalk. This makes it possible to take into account the instability of children's behavior.

[0036] The controller 100 may acquire statistical data regarding the traveling direction of pedestrians by time period. The controller 100 may change at least one of the first likelihood and the second likelihood based on the acquired statistical data. For example, as shown in FIG. 3, in the morning, the controller 100 may make the likelihood that pedestrian 60 will cross the crosswalk higher than the likelihood that pedestrian 61 will cross the crosswalk. On the other hand, as shown in FIG. 4, in the evening, the controller 100 may make the likelihood that pedestrian 61 will cross the crosswalk higher than the likelihood that pedestrian 60 will cross the crosswalk. This makes it possible to accurately predict whether at least one of the pedestrians will cross the crosswalk.

[0037] When the traveling direction of the pedestrian 60 (first pedestrian) and the pedestrian 61 (second pedestrian) is parallel to the traveling direction of the road, and when deceleration or stopping of either the pedestrian 60 or the pedestrian 61 is detected, or when a behavior change of the pedestrian 61 is not detected within a predetermined time, the controller 100 may lower at least one of the first likelihood and the second likelihood compared to when deceleration or stopping of either the pedestrian 60 or the pedestrian 61 is not detected, or when a behavior change of the pedestrian 61 is detected within a predetermined time. Consider a case where the pedestrians 60 and 61 are parallel to the traveling direction of the road as shown in FIG. 6 . In this scene, when deceleration or stopping of either the pedestrian 60 or the pedestrian 61 is detected, or when a behavior change of the pedestrian 61 is not detected within a predetermined time, it is estimated that there is a high possibility that the pedestrians 60 and 61 have just met an acquaintance across the road and exchanged a brief greeting. It is estimated that both (or at least one) of the pedestrians 60 and 61 are unlikely to cross the road to meet each other. Therefore, by lowering the likelihood, it becomes possible to accurately predict whether at least one of the pedestrians will cross the road. Note that, after detecting the deceleration or stoppage of either the pedestrian 60 or the pedestrian 61, if no change in the behavior of the pedestrian 61 is detected within a predetermined time, the controller 100 may lower the likelihood.

[0038] When the traveling directions of pedestrians 60 (first pedestrian) and 61 (second pedestrian) are parallel to the traveling direction of the road and change toward a crosswalk, controller 100 may increase the likelihood of one pedestrian whose traveling direction has changed compared to the likelihood of the other pedestrian whose traveling direction has not changed. For example, as shown in FIG. 7, when pedestrian 60 is detected, the traveling direction of pedestrian 60 may be parallel to the traveling direction of the road, but pedestrian 60 may later change his or her traveling direction toward a crosswalk in order to cross the crosswalk. This change indicates that there is a high possibility that pedestrian 60 will cross the crosswalk. Therefore, by calculating the likelihood based on this change, it becomes possible to accurately predict whether at least one of the pedestrians will cross the road.

[0039] The behavioral changes include changes in walking speed, gestures, hand movements, and facial orientation. By calculating the likelihood using such behavioral changes, it becomes possible to accurately predict whether at least one of the pedestrians will cross the road.

[0040] When at least one of the first likelihood and the second likelihood is greater than a predetermined value, the controller 100 accelerates the deceleration of the host vehicle 50 and stops the vehicle in front of the crosswalk. As a result, for example, when the host vehicle 50 is a vehicle equipped with an automatic driving function, the sense of discomfort felt by the occupants regarding the automatic driving is eliminated or reduced.

[0041] Each function described in the above embodiments may be implemented by one or more processing circuits. A processing circuit includes a programmed processing device, such as a processor including electrical circuitry. A processing circuit also includes devices, such as application specific integrated circuits (ASICs) or circuit components, arranged to perform the described functions.

[0042] Although the embodiments of the present invention have been described above, the descriptions and drawings that form part of this disclosure should not be understood to limit the present invention. Various alternative embodiments, examples, and operating techniques will become apparent to those skilled in the art from this disclosure.

[0043] In the above embodiment, a case where pedestrians 60 and 61 are detected on either side of a crosswalk has been described, but this is not limited to this. The present invention can also be applied to a case where pedestrians 60 and 61 are detected on both sides of a road other than a crosswalk, as shown in FIG. 8. [Explanation of symbols]

[0044] 1 object detection device, 2 vehicle position estimation device, 3 map database, 4 detection integration unit, 5 object tracking unit, 6 map position estimation unit, 10 behavior prediction unit, 11 pedestrian detection unit, 12 behavior change detection unit, 13 gaze detection unit, 14 attribute detection unit, 15 statistical data acquisition unit, 16 likelihood calculation unit, 20 vehicle control unit, 100 controller

Claims

1. A pedestrian crossing prediction method for a pedestrian crossing prediction device equipped with a sensor and a controller mounted on a host vehicle, comprising: The sensor Detecting a first pedestrian within a predetermined distance ahead of the vehicle; Detecting a second pedestrian on the other side of the road based on the position of the first pedestrian; The controller If a behavior change of the second pedestrian is detected within a predetermined time after the behavior change of the first pedestrian is detected by the sensor, at least one of a first likelihood that the first pedestrian will cross the road and a second likelihood that the second pedestrian will cross the road is increased compared to when a behavior change of the second pedestrian is not detected within the predetermined time, and it is determined whether the first pedestrian and the second pedestrian made eye contact before and after the timing when the behavior change of the second pedestrian occurred, and if it is determined that the eye contact did not occur, the increased likelihood is decreased. A pedestrian crossing prediction method.

2. The controller When it is determined that the gazes of the first and second walkers have met, at least one of the first likelihood and the second likelihood is changed based on a change in walking speed of the first walker and the second walker. The method for predicting pedestrian crossings according to claim 1 .

3. The controller increases the likelihood of one of the first and second walkers in which walking-related acceleration has been detected more than the likelihood of the other walker in which walking-related acceleration has not been detected. The method for predicting pedestrian crossings according to claim 2 .

4. The controller lowers the likelihood of one of the first and second pedestrians having detected deceleration related to walking compared to the likelihood of the other of the first and second pedestrians having not detected deceleration. The method for predicting pedestrian crossings according to claim 2 .

5. The controller increases the likelihood of one of the first and second pedestrians whose attribute is detected as a child more than the likelihood of the other of the first and second pedestrians whose attribute is not detected as a child. The pedestrian crossing prediction method according to any one of claims 1 to 4.

6. The controller Obtain statistical data on the direction of pedestrian movement by time of day, At least one of the first likelihood and the second likelihood is changed based on the acquired statistical data. The pedestrian crossing prediction method according to any one of claims 1 to 5.

7. When the traveling directions of the first pedestrian and the second pedestrian are parallel to the traveling direction of the road and a deceleration or stop of either the first pedestrian or the second pedestrian is detected, or a behavior change of the second pedestrian is not detected within the predetermined time, the controller lowers at least one of the first likelihood and the second likelihood compared to when a deceleration or stop of either the first pedestrian or the second pedestrian is not detected, or a behavior change of the second pedestrian is detected within the predetermined time. The pedestrian crossing prediction method according to any one of claims 1 to 6.

8. When the traveling directions of the first pedestrian and the second pedestrian are parallel to the traveling direction of the road and the traveling directions change toward a direction where a pedestrian crosses, the controller increases the likelihood of one of the pedestrians whose traveling direction has changed compared to the likelihood of the other pedestrian whose traveling direction has not changed. The pedestrian crossing prediction method according to any one of claims 1 to 7.

9. The behavioral changes include changes in walking speed, gestures, hand movements, and facial direction. The pedestrian crossing prediction method according to any one of claims 1 to 8.

10. The controller accelerates deceleration of the host vehicle when at least one of the first likelihood and the second likelihood is greater than a predetermined value. The pedestrian crossing prediction method according to any one of claims 1 to 9.

11. A pedestrian crossing prediction device comprising a sensor and a controller mounted on a vehicle, The sensor Detecting a first pedestrian within a predetermined distance ahead of the vehicle; Detecting a second pedestrian on the other side of the road based on the position of the first pedestrian; The controller If a behavior change of the second pedestrian is detected within a predetermined time after the behavior change of the first pedestrian is detected by the sensor, at least one of a first likelihood that the first pedestrian will cross the road and a second likelihood that the second pedestrian will cross the road is increased compared to when a behavior change of the second pedestrian is not detected within the predetermined time, and it is determined whether the first pedestrian and the second pedestrian made eye contact before and after the timing when the behavior change of the second pedestrian occurred, and if it is determined that the eye contact did not occur, the increased likelihood is decreased. A pedestrian crossing prediction device characterized by the above.

Images