Mobile body control device, mobile body control method, and program

Abstract

A control device for a moving body that moves by recognizing the periphery of the moving body is provided with: an object recognition unit that recognizes an object present in the periphery of the moving body on the basis of an observed value of a sensor provided in the moving body; and an object tracking unit that tracks the object recognized by the object recognition unit on the basis of time-series data of the observed values, the object tracking unit determining whether or not the object is a tracking target on the basis of the signal strength of the observed values and the variance of the observed values.

Description

TECHNICAL FIELD

[0001] The present application relates to a mobile body control device, a mobile body control method, and a program. BACKGROUND

[0002] In recent years, efforts toward realization of a low-carbon society or a decarbonized society are becoming active, and in vehicles as well, research and development related to electric mobile bodies that can move on both sidewalks and lanes are being conducted in order to reduce CO2 emissions and improve energy efficiency. For example, in Patent Literature 1, a technique is disclosed in which the influence of multiple paths is taken into account in an occupancy grid map in order to accurately estimate the surrounding environment of a mobile body from observation data obtained using a sensing device. Also, for example, in Patent Literature 2, a technique is disclosed in which the threshold value for determining whether or not an object target detected based on a sensor is a control target object target is set to different values in the case where the object target detected by a camera and an ultrasonic sonar is a fused object target and in the case where it is a non-fused object target.

[0003] PRIOR ART DOCUMENTS

[0004] PATENT LITERATURE

[0005] Patent Literature 1: Japanese Patent Application Publication No. 2017-166966

[0006] Patent Literature 2: Japanese Patent Application Publication No. 2020-190845 SUMMARY

[0007] PROBLEMS TO BE SOLVED BY THE INVENTION

[0008] In the above-described sensor fusion, object tracking using a Kalman filter is sometimes performed, but in the case where this is simply applied to the conventional technique, the noise is sometimes continuously tracked while the situation in which the noise is erroneously recognized as a tracking target continues.

[0009] The present application was completed in consideration of such a situation, and one of the objects thereof is to provide a mobile body control device, a mobile body control method, and a program that can suppress the situation in which noise is erroneously recognized as a tracking target from continuing in tracking of an object in the vicinity of a mobile body. Further, improvement in energy efficiency is facilitated.

[0010] MEANS FOR SOLVING THE PROBLEMS

[0011] The mobile body control device, the mobile body control method, and the program according to the present application adopt the following structure.

[0012] (1): One aspect of the present invention relates to a mobile body control device that recognizes a periphery of the mobile body and moves the mobile body, wherein the mobile body control device includes an object recognition unit that recognizes an object existing in the periphery of the mobile body based on an observation value of a sensor possessed by the mobile body, and an object tracking unit that tracks the object recognized by the object recognition unit based on time-series data of the observation value, the object tracking unit determining whether the object is a tracking target based on a signal strength of the observation value and a variance of the observation value.

[0013] (2): In the aspect of (1) above, the object tracking unit determines that the object is not a tracking target when a reliability of the observation value based on the signal strength is below a prescribed reference and the variance of the observation value is greater than a threshold value.

[0014] (3): In the aspect of (1) or (2) above, the object tracking unit starts monitoring of the variance of the observation value when the reliability changes from a state higher than the reference to a state below the reference, and determines that the object is no longer a tracking target when the variance of the observation value becomes greater than a threshold value.

[0015] (4): In any of the aspects of (1) to (3) above, the reliability is recognized based on an error between the signal strength of the observation value and an estimated result of the signal strength of the observation value, the object tracking unit determining that the object is not a tracking target when the error is greater than a first threshold value and the variance is greater than a second threshold value.

[0016] (5): In any of the aspects of (1) to (4) above, the object tracking unit learns the signal strength of the observation value using an autoencoder that temporarily compresses and restores the signal strength of the observation value, and compares a restoration error between the signal strength before compression by the autoencoder and the signal strength after restoration to the first threshold value.

[0017] (6): In the aspects of (1) to (5) above, the object tracking unit performs tracking of the object by applying the observation value to a Kalman filter, and performs initialization of the Kalman filter when it is determined that the object is not a tracking target.

[0018] (7): In the aspects of (1) to (6) above, the object tracking unit performs gating processing when the variance of the observation value is greater than a third threshold value, and performs update processing of the Kalman filter when there is an observation value that has passed the gating processing.

[0019] The mobile body control device according to claim 5.

[0020] (8) Another aspect of the present application relates to a control method of a mobile body that recognizes a periphery of the mobile body and moves, in which the control method of the mobile body causes a control device of the mobile body to execute: object recognition processing of recognizing an object existing in the periphery of the mobile body based on an observation value of a sensor possessed by the mobile body; and object tracking processing of tracking the object recognized by the object recognition processing based on time-series data of the observation value, in which, in the object tracking processing, it is determined whether the object is a tracking target based on a signal strength of the observation value and a variance of the observation value.

[0021] (9) Another aspect of the present application relates to a program in which a mobile body that recognizes a periphery of the mobile body and moves, the program causing a control device of the mobile body to execute: object recognition processing of recognizing an object existing in the periphery of the mobile body based on an observation value of a sensor possessed by the mobile body; and object tracking processing of tracking the object recognized in the object recognition processing based on time-series data of the observation value, in which, in the object tracking processing, it is determined whether the object is a tracking target based on a signal strength of the observation value and a variance of the observation value.

[0022] Effects of Invention

[0023] According to the aspects (1) to (9), in tracking of an object in the periphery of the mobile body, it is possible to suppress a situation in which noise is erroneously recognized as a tracking target. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 is a diagram that shows an example of a structure of a mobile body and a control device according to an embodiment.

[0025] Figure 2 is a perspective view of the mobile body as viewed from above.

[0026] Figure 3 is a flowchart that shows an example of a flow of processing of the control device tracking a dynamic object in the periphery of the mobile body.

[0027] Figure 4 is a diagram that shows an outline of the gating processing.

[0028] Figure 5 is a diagram that shows an outline of machine learning of the signal strength based on the autoencoder.

[0029] Figure 6 is a diagram that shows an example of a situation in which the object tracking section performs the loss determination. DETAILED DESCRIPTION

[0030] Embodiments of a mobile body control device, a mobile body control method, and a program of the present application will be described below with reference to the accompanying drawings. The mobile body can be a mobile body that moves in both a lane and a prescribed area different from the lane. The mobile body is sometimes referred to as a micro mobile body. An electric kick scooter is one type of micro mobile body. In addition, the mobile body can be a vehicle in which a passenger can ride, or an autonomous mobile body that can autonomously travel in a manner without a person. The latter autonomous mobile body is used, for example, for a purpose of transporting goods and the like. The prescribed area refers to, for example, a space in a house. In addition, the prescribed area can be a part or all of a road shoulder, a bicycle lane, a public open space, and the like, and can include all of a sidewalk, a road shoulder, a bicycle lane, a public open space, and the like. The control device can also be a control device that remotely controls the movement of the mobile body by transmitting an instruction from a location away from the mobile body using communication.

[0031] Figure 1 FIG. 1 is a diagram showing an example of the structure of a mobile body 1 and a control device 100 according to an embodiment. The mobile body 1 includes, for example, an external recognition device 10, a mobile body sensor 12, an operation member 14, an internal camera 16, a position measurement device 18, an acceleration sensor 20, a movement mechanism 30, a drive device 40, an external reporting device 50, a storage device 70, and the control device 100. Note that some of these structures can be omitted, which are not necessary for achieving the functions of the present application.

[0032] The external recognition device 10 is a device that sets the direction of travel of the mobile body 1 as a recognition range. The external recognition device 10 includes an external camera, a radar device, a LIDAR (Light Detection and Ranging), a sensor fusion device, and the like. The external recognition device 10 outputs information (an image, a position of an object, and the like) indicating a recognition result to the control device 100. The external recognition device 10 is not limited to a specific device as long as it can recognize the position of an object existing in the recognition range, but in this embodiment, the external recognition device 10 is provided as an example to output information of a wave reflected by a recognition target as an observation value. For example, the external recognition device 10 is a 3D sonar that uses a sound wave to detect an object.

[0033] The mobile body sensor 12 includes, for example, a speed sensor, a yaw rate (angular velocity) sensor, a direction sensor, and an operation amount detection sensor mounted on the operation member 14. The operation member 14 includes, for example, an operation member for instructing acceleration and deceleration (e.g., an accelerator pedal, a brake pedal), and an operation member for instructing steering (e.g., a steering wheel). In this case, the mobile body sensor 12 can include an accelerator opening degree sensor, a brake depression amount sensor, a steering torque sensor, and the like. The mobile body 1 can also have an operation member other than the above-described operation members (e.g., a rotary operation member other than a circular ring, a joystick, a button, and the like).

[0034] The interior camera 16 captures at least the head of the occupant of the mobile body 1 from the front. The interior camera 16 is a digital camera using a CCD (Charge Coupled Device), a CMOS (Complementary Metal Oxide Semiconductor), or the like. The interior camera 16 outputs the captured image to the control device 100.

[0035] The position measurement device 18 is a device that measures the position of the mobile body 1. The position measurement device 18 is, for example, a GNSS (Global Navigation Satellite System) receiver that determines the position of the mobile body 1 based on a signal received from a GNSS satellite and outputs the position as position information. Note that the position information of the mobile body 1 can also be estimated from the position of a Wi-Fi (registered trademark, also the same hereinafter) base station to which the communication device described later is connected.

[0036] The acceleration sensor 20 detects the acceleration of the mobile body 1 and outputs a signal corresponding to the detected acceleration to the control device 100. The acceleration sensor 20 detects the acceleration acting in the vertical direction (height direction) in addition to the horizontal direction of the mobile body 1.

[0037] The movement mechanism 30 is a mechanism for moving the mobile body 1. The movement mechanism 30 is, for example, a wheel group including a steering wheel and a drive wheel. In addition, the movement mechanism 30 can also be a leg for multi-legged walking.

[0038] The drive device 40 outputs a force to the movement mechanism 30 to move the mobile body 1. For example, the drive device 40 includes a motor that drives the drive wheel, a storage battery that stores power supplied to the motor, a steering device that adjusts the steering angle of the steering wheel, and the like. The drive device 40 can also have an internal combustion engine, a fuel cell, or the like as a drive force output unit or a power generation unit. In addition, the drive device 40 can also have a brake device based on frictional force, air resistance.

[0039] The external reporting device 50 is, for example, a lamp, a display device, a speaker, or the like provided to an outer panel of the mobile body 1 and used for reporting information to the outside of the mobile body 1. The external reporting device 50 can also perform different external reporting operations in accordance with the state of the mobile body 1 in motion, for example.

[0040] Figure 2 is a perspective view of the mobile body 1 as viewed from above. In the figure, FW is a steering wheel, RW is a drive wheel, SD is a steering device, MT is a motor, and BT is a storage battery. The steering device SD, the motor MT, and the storage battery BT are included in the drive device 40. In addition, AP is an accelerator pedal, BP is a brake pedal, WH is a steering wheel, SP is a speaker, and MC is a microphone. The illustrated mobile body 1 is a mobile body for one person to ride, and the occupant P is seated on a driver's seat DS and wears a seat belt SB. An arrow D1 is a direction of travel (a speed vector) of the mobile body 1. The external sensing device 10 is provided near a front end portion of the mobile body 1, and the internal camera 16 is provided at a position at which the head of the occupant P can be imaged from the front of the occupant P. In addition, the external reporting device 50 as a display device is provided near the front end portion of the mobile body 1.

[0041] Back Figure 1 The storage device 70 is, for example, a non-transitory storage device such as an HDD (Hard Disk Drive), a flash memory, a RAM (Random Access Memory), or the like. The map information 72, a program 74 executed by the control device 100, and the like are stored in the storage device 70. The storage device 70 is illustrated outside the block of the control device 100 in the figure, but the storage device 70 can be included in the control device 100.

[0042] [Control Device]

[0043] The control device 100 includes, for example, an object recognition unit 120, an object tracking unit 130, and a control unit 140. These are implemented by, for example, a hardware processor such as a CPU executing a program (software) 74. Some or all of these components can be implemented by hardware (including circuitry) such as an LSI (Large Scale Integration), an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), a GPU (Graphics Processing Unit), or the like, and can also be implemented by a combination of software and hardware. The program can be stored in advance in the storage device 70, or in a removable storage medium (non-transitory storage medium) such as a DVD or a CD-ROM, and installed in the storage device 70 by mounting the storage medium in a drive device.

[0044] The object recognition unit 120 recognizes an object present in the vicinity of the moving body 1 based on the output of the external perception device 10. The object includes some or all of a vehicle, a bicycle, a pedestrian, and the like, a moving body, a road division line, a step, a guardrail, a shoulder, a median strip, and the like, a road boundary, a road sign, a signboard, and the like, a structure provided on a road, a falling object, and the like, and an obstacle present on (falling on) a travel road. In addition, the object recognized by the object recognition unit 120 includes a static object that does not move from its place, and a dynamic object such as a person or another moving body. The object recognition unit 120 can recognize various states (position, speed, moving direction, and the like) of the object in addition to the presence of the object.

[0045] The object recognition unit 120 recognizes an object (another moving body, an obstacle, and the like) in the vicinity (mainly the area in the direction of travel) of the moving body 1, for example, by inputting an observation value of a sound wave acquired by the 3D sonar of the external perception device 10 to an object recognition model. Here, the object recognition model is, for example, a learned model obtained by learning using an algorithm of machine learning, and is learned so as to output information on the presence, position, and category of another moving body, and the like, by inputting an observation value of a sound wave acquired by the 3D sonar. Note that the category of another moving body can be estimated based on the size in an image, the intensity of a reflected wave received by the radar device of the external perception device 10, and the like, and can be estimated based on an image captured by the external camera of the external perception device 10. In addition, the object recognition unit 120 can combine these estimation results and perform recognition of an object. In addition, the object recognition unit 120 can acquire the speed of another moving body detected by the radar device using Doppler shift or the like, for example.

[0046] The object tracking unit 130 performs tracking processing of a dynamic object using a Kalman filter with respect to a dynamic object among the objects recognized by the object recognition unit 120. The object tracking unit 130 can estimate the position of a dynamic object that changes over time from the observation value of the outside perception device 10 (on the premise that there is an error) by using the Kalman filter. More specifically, the object tracking unit 130 estimates the current state based on the result of predicting the current state from the past state and the current observation value with respect to a dynamic object under tracking.

[0047] The control unit 140 controls the driving device 40, for example, in accordance with a set driving mode. The driving mode is set to any of a plurality of modes corresponding to the execution subject of steering and acceleration / deceleration. For example, the driving mode includes a first driving assistance mode in which a passenger performs a steering operation and automatic acceleration / deceleration control is performed, a second driving assistance mode in which a passenger performs an acceleration / deceleration operation and automatic steering control is performed, a manual driving mode in which a passenger performs a steering operation and an acceleration / deceleration operation, an automatic driving mode in which operation control and acceleration / deceleration control are automatically performed, and the like. The driving mode can be changed in accordance with the operation of the passenger, the control unit 140, and the like, and a mechanical switch that accepts a switching operation of the driving mode, a GUI (Graphical User Interface) switch provided on a touch panel, or the like can be provided to the mobile body 1.

[0048] For example, the control unit 140 sets a movement path that avoids collision with an object (static object and dynamic object) recognized by the object recognition unit 120 with respect to the mobile body 1 in accordance with the current position of the object, and controls the movement mechanism 30 and the driving device 40 of the mobile body 1 so as to move along the set movement path. The movement path can be appropriately determined in accordance with the driving mode, the set destination, and the like.

[0049] Figure 3 is a flowchart showing an example of the flow of processing in which the control device 100 tracks a dynamic object in the periphery of the mobile body 1. Here, processing of one cycle of processing that is repeatedly performed in the tracking processing will be described. That is, in actual tracking processing, Figure 3 the processing flow of is repeatedly performed. First, in the control device 100, the object tracking unit 130 determines whether or not the initialization flag of the Kalman filter is established (S101). As described above, Figure 3 the processing flow of is repeatedly performed, and therefore the object tracking unit 130 determines that the initialization flag is an initial value (for example, a value indicating a state in which the flag is not established) at the time of execution of S101, and thereafter determines the setting result of the initialization flag based on the last processing flow.

[0050] Here, if it is determined that the initialization flag has been established, the object tracking unit 130 initializes the Kalman filter (S102), obtains the current observation value from the external detection device 10 (S103), and uses the obtained observation value to perform prediction processing based on the Kalman filter (S104). On the other hand, if it is determined in S101 that the initialization flag has not been established, the object tracking unit 130 skips the initialization processing in S102 and proceeds to S103.

[0051] The initialization of the Kalman filter in S102 refers to returning the Kalman gain to its initial value. The initial value of the Kalman gain can be appropriately set based on the results of prior experiments, etc. The prediction processing of the Kalman filter in S104 is a process of predicting the future state of the tracked object based on the current state estimated about the tracked object. For example, the prediction processing of the Kalman filter assembles a motion model of the tracked object, and the object tracking unit 130 can obtain a predicted value of the state of the tracked object in the future (next cycle) by inputting the estimated result of the current state into the Kalman filter.

[0052] Next, the object tracking unit 130 calculates the variance of the observed values ​​of the tracked object and determines whether the value of the variance is greater than a predetermined threshold Vth1 (S105). If the variance is determined to be below the threshold Vth1, the object tracking unit 130 establishes an initialization flag (S106) and then terminates the series of processing steps. This is because, in this case, the observations of the tracked object are considered stable, and the reliability of the state prediction result predicted by the Kalman filter is high. By terminating after establishing the initialization flag, the Kalman filter can be initialized when starting the next processing cycle. The variance threshold Vth1 here is an example of a "third threshold".

[0053] On the other hand, if it is determined in S105 that the variance value is greater than the threshold Vth1, the object tracking unit 130 performs gating processing on the observations (S107) and determines whether there are any observations that have passed the gating processing (S108). The gating processing is the process of extracting the observations set as the tracking object from the acquired observation group. Figure 4 This is a diagram illustrating the general outline of the gating process. Figure 4 The example is obtained by plotting the observation group P11~P15 on the xy plane, which is viewed from above and encompasses the moving body 1 and its surroundings. The term "observations that have passed the gating process" refers to observations that were extracted as the tracking object during the gating process.

[0054] In this case, the object tracking section 130 performs a gating process in which observation values Pll to P13 included in a prescribed detection range Rl of the observation value group Pl l to P15 existing in the range of the field of view when viewed from the viewpoint position of the mobile body 1 (the viewpoint position of the outside recognizing device 10) are extracted. Figure 4 The example of FIG. 10 shows a case where the detection range Rl is defined by a rectangle having the front direction of the mobile body 1 as the long side direction, but this is an example, and the detection range Rl can be arbitrarily set according to the desired detection sensitivity, the directivity of the detection, and the like. By such a gating process, the observation values of the processing target are narrowed in range, whereby it is possible to suppress a case where the real-time performance is reduced due to a large processing amount.

[0055] Returning to Figure 3 Next, in a case where the object tracking section 130 determines that there is no observation value that has passed the gating process in S108, the series of processing procedures ends. This is because "there is no observation value that has passed the gating process" means that there is no observation value of the tracking object (processing target). On the other hand, in a case where the object tracking section 130 determines that there is an observation value that has passed the gating process in S108, the object tracking section 130 performs update processing of the Kalman filter (S109).

[0056] As described above, the Kalman filter estimates the true state of the tracking object in the current period on the basis of the observation value of the state of the tracking object in the current period and the Kalman gain, and the state of the tracking object in the current period predicted from the state estimation result in the previous period. More specifically, the Kalman filter inputs the observation value and the observation error every period, updates the Kalman gain in accordance with the observation error and the estimation error in the current period predicted in the filtering calculation in the previous period (which is the unreliability of the estimation, hereinafter referred to as "predicted estimation error"), and estimates the true value of the state value in the current period in accordance with the updated Kalman gain and the state value in the current period predicted in the filtering calculation in the previous period (hereinafter referred to as "predicted state value"). In addition, the Kalman filter calculates the estimation error of the true value, and outputs the estimation value and the estimation error as the estimation result in the current period. The update processing of S109 includes such update processing of the Kalman gain, and processing of the state estimation of the tracking object using the updated Kalman gain.

[0057] Note that the Kalman filter calculates the state value (predicted state value) and the estimation error (predicted estimation error) in the next period in accordance with the result of the state estimation in the current period, and sets the prediction result as the input of the update processing in the next period.

[0058] Next, the object tracking unit 130 performs learning of the signal intensity based on machine learning with respect to the observation value of the current cycle (S110). Here, an auto-encoder is described as an algorithm of machine learning, but the algorithm used is not limited to this. The auto-encoder is one of methods of unsupervised machine learning using a neural network, and performs unsupervised learning for the purpose of outputting data identical to the input data.

[0059] Figure 5 is a diagram illustrating an outline of machine learning of the signal intensity based on the auto-encoder. The auto-encoder is a structure that compresses input data to temporarily reduce the number of dimensions and restores the number of dimensions to output, and thus can form a network that "extracts important information and efficiently generates original data using the extraction result" for data restoration. In addition, the auto-encoder performs unsupervised learning as described above, and thus can perform learning of the signal intensity online by cumulatively inputting the signal intensity of the observation data of each cycle.

[0060] In addition, it is also possible to detect a decrease in reliability of the signal intensity not by the auto-encoder but by directly comparing the signal intensity with a threshold value, but in this case, it is necessary to set the threshold value based on rules in advance, but it is difficult to set rules considering all situations with respect to the signal intensity in advance, and becomes a high-cost operation. In contrast, in the case of using the auto-encoder, the restoration error between the input and output of the auto-encoder is compared with the threshold value, and if the restoration error is larger than the threshold value, it can be set as abnormal, and if the restoration error is equal to or smaller than the threshold value, it can be determined as normal, and thus it is possible to more simply and in a form of coping with more situations to detect a decrease in reliability of the signal intensity.

[0061] Return Figure 3 Next, the object tracking unit 130 inputs the signal intensity of the observation value of the current cycle to the auto-encoder learned in S110, and acquires, as an output thereof, the signal intensity obtained by temporarily compressing and restoring the signal intensity inside the auto-encoder. The object tracking unit 130 calculates the restoration error of the output signal intensity with respect to the input signal intensity (S111), and determines whether the restoration error is larger than a prescribed threshold value Eth (S112). Here, in the case where it is determined that the restoration error is equal to or smaller than the threshold value Eth, the object tracking unit 130 ends the series of processing procedures. This is based on the insight that, in the case where noise is erroneously detected as a tracking object, the deviation of the signal intensity becomes large, and the restoration error easily becomes large. The threshold value Eth of the restoration error here is an example of a "first threshold value".

[0062] On the other hand, if in S112 it is determined that the restoration error is greater than the threshold Eth, the object tracking unit 130 calculates the variance of the observations obtained during the period from the present to a predetermined point in the past, and determines whether the variance is greater than the predetermined threshold Vth2 (S113). Here, if it is determined that the variance is less than or equal to the threshold Vth2, the object tracking unit 130 terminates the series of processing steps. This is based on the insight that even when the deviation in signal strength is large, the possibility that it is not noise is high when the deviation in the observations is small. The variance threshold Vth2 here is an example of a "second threshold".

[0063] On the other hand, if in S113 the variance of the observed value is determined to be greater than the threshold Vth2, the object tracking unit 130 determines that the object involved in the observed value is noise (S114: loss determination). In this case, the object tracking unit 130 establishes an initialization flag (S115) and then terminates a series of processing flows. Here, by establishing the initialization flag, the Kalman filter is initialized at the beginning of the next cycle, and then, when a tracking object is generated, prediction processing and update processing are repeatedly performed, thereby updating the Kalman filter at any time.

[0064] It should be noted that in the above processing flow, the object tracking unit 130 can set multiple objects (including noise) as tracking objects. That is, the object tracking unit 130 can manage the Kalman filter and initialization flag for each tracking object. In addition, the object tracking unit 130 can learn the signal strength of the observed values ​​for each tracking object. Furthermore, the object tracking unit 130 can determine the restoration error and variance for each tracking object.

[0065] Figure 6 This diagram illustrates an example of the object tracking unit 130 performing a loss determination. Figure 6 The upper chart represents the monitoring status of signal strength, and the lower chart represents the monitoring status of the variance of the observations. In this example, the object tracking unit 130 sequentially learns and monitors the signal strength of the observations input in a time series, and identifies the signal strength of the observations during the period t0~t1 as the signal strength in cases where a person has been detected. It should be noted that, as described above, this is identified by determining that the signal strength recovery error is below the threshold Eth, that is, this corresponds to a situation where the reliability of the observations is higher than a predetermined benchmark.

[0066] Next, the object tracking unit 130 detects that the signal strength of the observation value has become the signal strength in the case where noise is detected at time t2. This is recognized by determining that the recovery error of the signal strength is larger than the threshold value Eth as described above, that is, this corresponds to the case where the reliability of the observation value is below the prescribed reference. The object tracking unit 130 determines that the reliability of the observation value is likely to have decreased in accordance with the change in the signal strength, and starts monitoring of the variance of the observation value from this time point t2.

[0067] Figure 6 The example of FIG. 10 shows a situation in which the tracking object becomes noise from the vicinity of time t2 at which the signal strength has greatly decreased, and the variance of the observation value gradually increases. In this example, the object tracking unit 130, after starting the variance monitoring at time t2, detects that the variance becomes larger than the threshold value Vth2 for the loss determination at time t3, and performs the loss determination.

[0068] According to the present embodiment described above, it is possible to accurately recognize that the likelihood that noise becomes the tracking object has increased by monitoring the change in the signal strength of the observation value, and it is possible to perform the loss determination based on the variance of the observation value on this basis. Therefore, it is possible to quickly recognize the situation in which noise is erroneously recognized as the tracking object, and to suppress the continuation of this situation.

[0069] In addition, according to the present embodiment, the monitoring of the variance is started from the time when it is recognized that the likelihood that noise becomes the tracking object has increased, and thus in the case where the processing load of the variance monitoring is high, it is possible to reduce the hardware resources required for the variance monitoring, and to reduce the processing load of the control device 100.

[0070] The embodiment described above can be expressed as follows.

[0071] A control device of a mobile body that identifies a periphery of the mobile body and moves,

[0072] The control device of the mobile body includes:

[0073] a storage medium that stores computer-readable instructions; and

[0074] a processor connected to the storage medium,

[0075] the processor executing the computer-readable instructions to:

[0076] An object recognition process recognizes an object existing in the periphery of the mobile body on the basis of an observation value of a sensor possessed by the mobile body; and

[0077] An object tracking process tracks the object recognized in the object recognition process on the basis of time-series data of the observation value,

[0078] In the object tracking process, it is determined whether the object is a tracking target on the basis of a signal strength of the observation value and a variance of the observation value.

[0079] The above describes the specific embodiments of the present application using the embodiments, but the present application is not at all limited by such embodiments, and various modifications and substitutions can be made within the scope of the gist of the present application.

[0080] Explanation of Reference Numerals

[0081] 1…mobile body, 10…outside sensing device, 12…mobile body sensor, 14…operation member, 16…internal camera, 18…positioning device, 20…acceleration sensor, 30…moving mechanism, 40…drive device, 50…external reporting device, 70…storage device, 72…map information, 74…program, 100…control device, 120…object recognition section, 130…object tracking section, 140…control section.

Claims

1. A mobile body control device that identifies a periphery of the mobile body and moves, wherein the mobile body control device includes: an object identification unit that identifies an object existing in the periphery of the mobile body based on an observation value of a sensor possessed by the mobile body; and an object tracking unit that tracks the object identified by the object identification unit based on time-series data of the observation value, the object tracking unit determines whether the object is a tracking target based on a signal strength of the observation value and a variance of the observation value.

2. The mobile body control device according to claim 1, wherein the object tracking unit determines that the object is not a tracking target when a reliability of the observation value based on the signal strength is below a prescribed reference and the variance of the observation value is larger than a threshold value.

3. The mobile body control device according to claim 2, wherein the object tracking unit starts monitoring of the variance of the observation value when the reliability changes from a state higher than the reference to a state below the reference, and determines that the object is no longer a tracking target when the variance of the observation value becomes larger than the threshold value.

4. The mobile body control device according to claim 2, wherein the reliability is identified based on an error between the signal strength of the observation value and an estimation result of the signal strength of the observation value, and the object tracking unit determines that the object is not a tracking target when the error is larger than a first threshold value and the variance is larger than a second threshold value.

5. The mobile body control device according to claim 4, wherein the object tracking unit learns the signal strength of the observation value using an autoencoder that temporarily compresses and restores the signal strength of the observation value, and compares a restoration error between the signal strength before compression by the autoencoder and the signal strength after restoration to the first threshold value.

6. The mobile body control device according to claim 1, wherein the object tracking unit performs tracking of the object by applying the observation value to a Kalman filter, and performs initialization of the Kalman filter when it is determined that the object is not a tracking target.

7. The mobile body control device according to claim 6, wherein the object tracking unit performs a gating process when the variance of the observation value is larger than a third threshold value, and performs an update process of the Kalman filter when there is an observation value that passes the gating process.

8. A mobile body control method that identifies a periphery of the mobile body and moves, wherein the mobile body control method causes a mobile body control device to perform the following processes: an object identification process that identifies an object existing in the periphery of the mobile body based on an observation value of a sensor possessed by the mobile body; and an object tracking process that tracks the object identified in the object identification process based on time-series data of the observation value, ​ ​ In the object tracking process, it is determined whether the object is a tracking target based on a signal strength of the observation value and a variance of the observation value.

9. A program, wherein, The mobile body identifies a periphery of the mobile body and moves, The program causes a control device of the mobile body to execute the following processes: An object recognition process that recognizes an object existing in a periphery of the mobile body based on an observation value of a sensor possessed by the mobile body; and An object tracking process that tracks the object recognized in the object recognition process based on time-series data of the observation value, In the object tracking process, it is determined whether the object is a tracking target based on a signal strength of the observation value and a variance of the observation value.

Images