processing system, processing device, processing method, processing program

Abstract

A processing system, a processing device, a processing method, and a processing program are provided. A processor of a processing system that performs identification-related processing associated with identification of a target mobile body in a host mobile body is configured to perform processing of: acquiring a probability distribution (Dp) of presence of the target mobile body with respect to a distance (Lh) from the host mobile body based on a travel area (Ad) of the host mobile body; acquiring an identification rate of identifying the target mobile body for each of a plurality of identification models (Mn); and fusing identification data of each of the identification models (Mn) based on a fusion rate (ωn) based on an identification score (Sn) associated with each identification model Mn and the identification rate and the probability distribution (Dp).

Description

[0001] Cross-reference of related applications

[0002] This application is based on Japanese Patent Application No. 2021-115163, filed on July 12, 2021, by reference in its entirety to the contents of the base application. Technical Field

[0003] This disclosure relates to techniques for performing identification association processing associated with the identification of a target mobile entity in a master mobile entity. Background Technology

[0004] The technology disclosed in Patent Document 1 uses multiple dictionaries as recognition models in order to identify pedestrians as target moving bodies in a vehicle that is the main moving body.

[0005] Patent Document 1: Japanese Patent Application Publication No. 2008-20951

[0006] The technology disclosed in Patent Document 1 searches for and selectively uses a dictionary with the lowest resolution in order to reduce computation time. However, since the required recognition performance of the dictionary varies depending on the vehicle's driving area, there is a concern that recognition accuracy cannot be guaranteed solely by using a dictionary based on the lowest resolution. Summary of the Invention

[0007] The present disclosure addresses the following issues: A processing system for ensuring the accuracy of target motion object identification in a main moving body; a processing apparatus for ensuring the accuracy of target motion object identification in a main moving body; a processing method for ensuring the accuracy of target motion object identification in a main moving body; and a further processing procedure for ensuring the accuracy of target motion object identification in a main moving body.

[0008] The technical means of this disclosure used to solve the problem will be described below.

[0009] The first aspect of this disclosure is a processing system having a processor that performs identification association processing associated with the identification of a target moving body in a main moving body.

[0010] The processor is configured to perform the following processes:

[0011] The probability distribution of the existence of a target mobile object relative to the distance from the main mobile object is obtained based on the driving area of ​​the main mobile object.

[0012] For each of the multiple recognition models, obtain the recognition rate of the target moving object; and

[0013] The recognition data of each recognition model are fused according to the fusion rate based on the recognition score, wherein the recognition score is related to the recognition rate and probability distribution of each recognition model.

[0014] The second aspect of this disclosure is a processing apparatus having a processor configured to be mounted on a main mobile body, and performing identification association processing associated with the identification of a target mobile body within the main mobile body.

[0015] The processor is configured to perform the following processes:

[0016] The probability distribution of the existence of a target mobile object relative to the distance from the main mobile object is obtained based on the driving area of ​​the main mobile object.

[0017] For each of the above recognition models, obtain the recognition rate of the target moving object; and

[0018] The recognition data of each recognition model are fused according to the fusion rate based on the recognition score, where the recognition score is related to the recognition rate and probability distribution of each recognition model.

[0019] The third aspect of this disclosure is a processing method executed by a processor for performing identification association processing associated with the identification of a target mobile body in a main mobile body, comprising:

[0020] The probability distribution of the existence of a target mobile object relative to the distance from the main mobile object is obtained based on the driving area of ​​the main mobile object.

[0021] For each of the above recognition models, obtain the recognition rate of the target moving object; and

[0022] The recognition data of each recognition model are fused according to the fusion rate based on the recognition score, where the recognition score is related to the recognition rate and probability distribution of each recognition model.

[0023] The fourth aspect of this disclosure is a processing program stored in a storage medium for performing identification association processing associated with the identification of a target mobile body in a main mobile body, and containing instructions for execution by a processor.

[0024] The instructions include:

[0025] The probability distribution of the existence of a target mobile object relative to the distance from the main mobile object is obtained based on the driving area of ​​the main mobile object.

[0026] For each of the above recognition models, obtain the recognition rate of the target moving object; and

[0027] The recognition data of each recognition model are fused according to the fusion rate based on the recognition score, where the recognition score is related to the recognition rate and probability distribution of each recognition model.

[0028] According to these first to fourth methods, a probability distribution of the presence of a target mobile object relative to the distance from the main mobile object is obtained based on the driving area of ​​the main mobile object. Therefore, in the first to fourth methods, the recognition data of each recognition model is fused according to a fusion rate based on a recognition score related to the recognition rate and the probability distribution thus obtained for each of the multiple recognition models for identifying the target mobile object. Thus, even if the required recognition performance changes according to the driving area, appropriate fusion can be achieved by following the fusion rate, which matches the recognition score reflected by the probability distribution. Therefore, the accuracy of target mobile object recognition within the main mobile object can be ensured. Attached Figure Description

[0029] Figure 1 It is a block diagram showing the overall structure of one implementation method.

[0030] Figure 2 This is a schematic diagram illustrating the driving environment of a main vehicle using one implementation method.

[0031] Figure 3 It is a block diagram representing the functional structure of a processing system in one implementation.

[0032] Figure 4 It is a flowchart illustrating the processing flow of one implementation method.

[0033] Figure 5 It is a graph used to illustrate the probability distribution of one implementation method.

[0034] Figure 6 It is a graph used to illustrate the velocity distribution of one implementation method.

[0035] Figure 7 It is a graph used to illustrate the acceleration distribution of one implementation.

[0036] Figure 8 This is a schematic diagram illustrating a track intersection scenario in one implementation method.

[0037] Figure 9 This is a schematic diagram illustrating a track intersection scenario in one implementation method.

[0038] Figure 10 This is a schematic diagram illustrating a track intersection scenario in one implementation method.

[0039] Figure 11This is a schematic diagram illustrating a track intersection scenario in one implementation method.

[0040] Figure 12 This is a schematic diagram illustrating a track intersection scenario in one implementation method.

[0041] Figure 13 This is a characteristic table representing the critical gap of one implementation method.

[0042] Figure 14 This is a block diagram representing multiple recognition models in one implementation method.

[0043] Figure 15 This is a schematic diagram used to illustrate the recognition rate of one implementation method.

[0044] Figure 16 This is a feature table used to describe the recognition rate of one implementation method.

[0045] Figure 17 It is a feature table used to describe the recognition score of one implementation.

[0046] Figure 18 This is a feature table used to explain the optimization calculation of the fusion rate of one implementation.

[0047] Figure 19 This is a schematic diagram used to illustrate the limitations of the operating design area of ​​one implementation method. Detailed Implementation

[0048] Hereinafter, an embodiment of the present disclosure will be described with reference to the accompanying drawings.

[0049] Figure 1 The processing system 1 shown in one embodiment executes and acts as the main moving body Figure 2 The identification association processing associated with the identification of the target moving body 3 in the main vehicle 2 shown. From the viewpoint centered on the main vehicle 2, the main vehicle 2 can also be referred to as the ego-vehicle. The main vehicle 2 is a moving body, such as a car, capable of traveling on a road with passengers. From the viewpoint centered on the main vehicle 2, the target moving body 3 can also be referred to as other road users. The target moving body 3 includes, for example, at least one of the following: a car, a motorcycle, a bicycle, an autonomous robot, a pedestrian, and an animal.

[0050] In the main vehicle 2, an autonomous driving mode is provided, which is categorized based on the degree of manual intervention by the occupants during the driving task. The autonomous driving mode can also be achieved through autonomous driving control where the system performs all driving tasks, such as conditional driving automation, high driving automation, or full driving automation. Alternatively, the autonomous driving mode can be achieved through high-level driving assistance control, such as driver assistance or partial driving automation, where the occupants perform some or all of the driving tasks. The autonomous driving mode can also be achieved through any one of these autonomous driving controls and high-level driving assistance controls, a combination of them, or a switching between them.

[0051] Carrying in main vehicle 2 Figure 1 The sensor system 4, communication system 5, information prompting system 6, and map database 7 are shown. The sensor system 4 acquires sensor information that can be used by the processing system 1 by detecting the external and internal boundaries of the main vehicle 2. For this purpose, the sensor system 4 is composed of an external sensor 40 and an internal boundary sensor 42.

[0052] The external sensor 40 acquires external information as sensor information from the external environment surrounding the main vehicle 2. The external sensor 40 can also acquire external information by detecting objects present outside the main vehicle 2. The object detection type external sensor 40 is, for example, at least one of a camera, LiDAR (Light Detection and Ranging / Laser Imaging Detection and Ranging), radar, and sonar.

[0053] The interior boundary sensor 42 acquires interior boundary information that can be utilized by the processing system 1 from the interior boundary, which constitutes the interior environment of the main vehicle 2. The interior boundary sensor 42 can also acquire interior boundary information by detecting specific motion physical quantities within the interior boundary of the main vehicle 2. Physical quantity detection type interior boundary sensor 42 is, for example, at least one of a driving speed sensor, an acceleration sensor, and a gyroscope sensor. The interior boundary sensor 42 can also acquire interior boundary information by detecting specific states of occupants within the interior boundary of the main vehicle 2. Occupant detection type interior boundary sensor 42 is, for example, at least one of a driver status monitor (registered trademark), a bio-sensor, a seating sensor, an actuator sensor, and an in-vehicle equipment sensor.

[0054] Communication system 5 acquires communication information that can be used by processing system 1 via wireless communication. Communication system 5 can also receive positioning signals from GNSS (Global Navigation Satellite System) satellites located outside the main vehicle 2. Positioning-type communication system 5 may be, for example, a GNSS receiver. Communication system 5 can also transmit and receive communication signals with V2X systems located outside the main vehicle 2. V2X-type communication system 5 may be, for example, at least one of DSRC (Dedicated Short Range Communications) communicators and cellular V2X (C-V2X) communicators. Communication system 5 can also transmit and receive communication signals with terminals located inside the main vehicle 2. Terminal communication-type communication system 5 may be, for example, at least one of Bluetooth devices, Wi-Fi devices, and infrared communication devices.

[0055] Information prompting system 6 prompts the occupants of the main vehicle 2 with report information. Information prompting system 6 can also prompt report information by stimulating the occupants' vision. Visual stimulation type information prompting system 6 is, for example, at least one of HUD (Head-Up Display), MFD (Multi-Function Display), instrument cluster, navigation unit, and luminous unit. Information prompting system 6 can also prompt report information by stimulating the occupants' hearing. Auditory stimulation type information prompting system 6 is, for example, at least one of speaker, buzzer, and vibration unit.

[0056] Map database 7 stores map information that can be used by processing system 1. Map database 7 may be constructed using at least one non-transitorytangible storage medium, such as semiconductor memory, magnetic media, or optical media. Map database 7 may also be a database of a locator that estimates its own state quantities, including the position of the main vehicle 2. Map database 7 may also be a database of a navigation unit that navigates the driving path of the main vehicle 2. Map database 7 may also be composed of a combination of multiple types of such databases.

[0057] Map database 7 acquires and stores the latest map information, for example, through communication with an external center via a V2X-type communication system 5. Here, the map information is digitized in two or three dimensions to represent the driving environment of the main vehicle 2. In particular, as three-dimensional map data, high-precision digital map data can be used. The map information may also include, for example, road information indicating at least one of the following: the location, shape, and road surface condition of the road itself. The map information may also include, for example, identification information indicating at least one of the location and shape of road signs and markings attached to the road. The map information may also include, for example, structural information indicating at least one of the location and shape of buildings and traffic lights facing the road.

[0058] The processing system 1 is connected to the sensor system 4, the communication system 5, the information prompting system 6, and the map database 7, for example, via at least one of the following: a LAN (Local Area Network) line, a wiring harness, an internal bus, and a wireless communication line. The processing system 1 comprises at least one dedicated computer.

[0059] The dedicated computer constituting processing system 1 can also be a driving control ECU (Electronic Control Unit) that controls the driving of the main vehicle 2. The dedicated computer constituting processing system 1 can also be a navigation ECU that navigates the driving path of the main vehicle 2. The dedicated computer constituting processing system 1 can also be a locator ECU that estimates the self-state of the main vehicle 2. The dedicated computer constituting processing system 1 can also be an actuator ECU that controls the driving actuators of the main vehicle 2. The dedicated computer constituting processing system 1 can also be an HCU (Hman Machine Interface Control Unit) that controls the information display system 6 in the main vehicle 2. The dedicated computer constituting processing system 1 can also be a computer outside the main vehicle 2, such as an external center or mobile terminal capable of communicating via a V2X-type communication system 5.

[0060] The dedicated computer constituting the processing system 1 has at least one memory 10 and at least one processor 12. The memory 10 is a non-transitory tangible storage medium that non-transitorily stores computer-readable programs and data, such as semiconductor memory, magnetic media, and optical media. The processor 12 includes, for example, at least one of the following as its core: CPU (Central Processing Unit), GPU (Graphics Processing Unit), RISC (Reduced Instruction Set Computer) – CPU, DFP (Data Flow Processor), and GSP (Graphics Streaming Processor).

[0061] In processing system 1, processor 12 executes multiple instructions contained in a processing program stored in memory 10 for performing identification and association processing in main vehicle 2. Thus, processing system 1 constructs multiple functional modules for performing identification and association processing in main vehicle 2. For example... Figure 3 As shown, the multiple functional modules constructed in the processing system 1 include a distribution acquisition block 100, a recognition rate acquisition block 110, a fusion block 120, and a restriction block 130.

[0062] Through the collaboration of these blocks 100 and 110, according to Figure 4 The processing flow shown executes a processing method for the identification and association processing performed by the processing system 1 in the main vehicle 2. This processing flow is repeatedly executed during the startup of the main vehicle 2. Furthermore, each "S" in this processing flow represents a multiple step executed by multiple instructions contained in the processing program.

[0063] In processing step S101, the distribution acquisition block 100 determines the driving area Ad (referencing) of the main vehicle 2 by judging the driving scene including the main vehicle 2's own position based on the information acquired or stored by the sensor system 4, communication system 5, and map database 7. Figure 2 At this point, the driving area Ad is defined as the area within the distance range of the target moving body 3 that the main vehicle 2 needs to identify, which is located to the left and right in the lateral direction and to the front and back in the longitudinal direction of the main vehicle 2.

[0064] In process S102, such as Figure 5 As shown, the distribution acquisition block 100 obtains the probability distribution Dp of the existence of a target moving body 3 relative to the distance Lh from the main vehicle 2 based on the driving area Ad of the main vehicle 2.

[0065] Specifically, the distribution acquisition block 100 of S102 acquires information representing the velocity vm of the moving body in the driving area Ad based on the information acquired or stored by the sensor system 4, the communication system 5, and the map database 7. Figure 6 The velocity distribution Dv is shown as either continuously dispersed or discontinuously dispersed (i.e., histogram dispersion). Accompanying this, the distribution acquisition block 100 in S102 acquires, based on acquired or stored information from the sensor system 4, communication system 5, and map database 7, the acceleration am of the moving body within the driving area Ad. Figure 7 The acceleration distribution Da is shown as either continuously dispersed or discontinuously dispersed. These velocity distributions Dv and acceleration distributions Da provide an overall proportion or quantity as the frequency at which the velocity vm or acceleration am of the moving body is acquired. Therefore, the velocity distribution Dv and acceleration distribution Da can be obtained using data specific to each driving area Ad, such as the latest traffic information accumulated at the outer center.

[0066] S102's distribution acquisition block 100 acquires the critical gap tc of the moving body in the driving area Ad and the identification time ts of the main vehicle 2 in the driving area Ad, as shown in Equation 1 and Figure 5 The larger side is shown as the upper limit time tm.

[0067] [Formula 1]

[0068] tm = max(tc, ts)

[0069] Here, the critical gap tc is, for example, as follows: Figures 8-12 As shown, this refers to the gap in scenarios where multiple mobile vehicles' tracks intersect, such as entering, merging, or changing lanes at intersections, where the probability of a mobile vehicle on the track-changing side performing the track change is equal to the probability of delaying the track change. Additionally, in Figures 8-12 In this context, the moving body on the side that causes the change in the travel track is depicted as the main vehicle 2, and the moving body on the side that forms the gap is depicted as the target moving body 3.

[0070] Therefore, for example, the latest gap acceptance information accumulated at the external center can be used, such as... Figure 13 The critical gap tc is obtained as illustrated. Alternatively, when no track crossing scenario is assumed in the main vehicle 2, the critical gap tc can be forced to a value of 0.

[0071] On the other hand, the recognition time ts refers to the maximum time required for the main vehicle 2 to recognize the target moving body 3 within the travel area Ad. Therefore, for example, the recognition time ts can be obtained using path planning information from the main vehicle 2. Furthermore, for example, in a track-intersecting scenario, the recognition time ts can be set as the time from the current position of the main vehicle 2 until the start of track changes.

[0072] like Figure 5 As shown, in S102, the distribution acquisition block 100 acquires a probability distribution Dp corresponding to the driving area Ad within the travel distance range ΔL of the main vehicle 2 during the period from the current time to the upper limit time tm. Here, the probability distribution Dp represents the overall proportion or number of target moving objects 3 existing relative to the distance Lh from the main vehicle 2. That is, the probability distribution Dp provides the overall proportion or number of dispersions as the probability of the existence of target moving objects 3 depending on the distance Lh. Therefore, the probability distribution Dp is acquired by synthesizing the velocity distribution Dv and the acceleration distribution Da to satisfy Equation 2 with the upper limit time tm as the variable.

[0073] [Equation 2]

[0074]

[0075] exist Figure 4 In S103 of the processing flow shown, the recognition rate acquisition block 110 acquires the recognition rate of the target moving body 3 in the driving area Ad for each of the multiple recognition models Mn. Here, as Figure 14 As shown, each recognition model Mn is stored in memory 10 as a rule model (i.e., recognition logic) or machine learning model used to identify the target moving body 3 based on the driving scene in the driving area Ad. Additionally, the suffix n in the reference numeral Mn refers to an integer from 1 to N, serving as an index for distinguishing the N recognition models Mn respectively.

[0076] Specifically, each recognition model Mn pairs as follows Figure 15 As shown, the driving area Ad is divided into multiple rectangular grid-like regions (i.e., also called squares) Am in the left and right sides laterally and front and back longitudinally of the main vehicle 2, respectively, with a specified recognition rate for each grid region. As long as no gaps are created between these grid regions Am, the shape of each grid region Am can be set to a shape other than a rectangular grid. Furthermore, in Figure 15 In the diagram, the values ​​within each grid region Am represent the recognition rate.

[0077] like Figure 16As shown, in each recognition model Mn of this embodiment, the recognition rate specified for each grid region Am includes at least one of the consistency between the real world and the recognition (TP, TN) and the inconsistency between the real world and the recognition (FP, FN). Here, consistency (TP) is the recognition success rate when the recognition result is consistent with the existence of the target moving object 3 in the real world. Consistency (TN) is the recognition success rate when the recognition result is consistent with the non-existence of the target moving object 3 in the real world. Inconsistency (FP) is the false recognition rate when the recognition result is inconsistent with the existence of the target moving object 3 in the real world. Inconsistency (FN) is the false recognition rate when the recognition result is inconsistent with the non-existence of the target moving object 3 in the real world.

[0078] exist Figure 4 In S104 of the processing flow shown, the fusion block 120 acquires the recognition data of each recognition model Mn based on the acquired or stored information from the sensor system 4, the communication system 5, and the map database 7. The recognition data of each recognition model Mn is acquired as data related to the target moving body 3, and includes at least one of the following: position, speed, shape, type, and recognition confidence (i.e., recognition reliability).

[0079] In processing step S105, the fusion block 120 fuses the recognition data of each recognition model Mn according to the fusion rate ωn, wherein the fusion rate ωn is based on, for example... Figure 17 As shown, the recognition score Sn is associated with each recognition model Mn and the recognition rate and probability distribution Dp. Additionally, the suffix n in the reference numerals Sn and ωn refers to an integer from 1 to N, serving as an index to distinguish the recognition score Sn and fusion rate ωn corresponding to the N recognition models Mn respectively.

[0080] Specifically, the fusion block 120 of S105 obtains the recognition score Sn of each recognition model Mn by correcting the recognition rate of each recognition model Mn using the probability distribution Dp. At this time, the probability of the existence of the target moving body 3, which depends on the distance Lh in the probability distribution Dp, is multiplied by... Figure 17 The recognition rate of each recognition model Mn is used to obtain the corrected recognition score Sn.

[0081] Therefore, in multiple grid regions Am within the travel distance range ΔL of probability distribution Dp from the main vehicle 2 to the region center, the acquisition of a recognition score Sn, which corrects the recognition rate of each recognition model Mn by multiplying this distance Lh by the existence probability, is performed. Here, if the recognition rate of each recognition model Mn includes multiple categories of consistency degree TP, TN, and inconsistency degree FP, FN, the existence probability is multiplied by each of these categories. On the other hand, if the recognition rate of each recognition model Mn includes one category of consistency degree TP, TN, and inconsistency degree FP, FN, the existence probability is multiplied by that category. However, in at least one grid region Am outside the travel distance range ΔL of probability distribution Dp from the main vehicle 2 to the region center, the acquisition of the recognition score Sn in each recognition model Mn is stopped.

[0082] The fusion block 120 of S105 optimizes the fusion rate ωn of each recognition model Mn based on the recognition score Sn obtained for each recognition model Mn. At this time, the recognition score Sn of each grid region Am within the movement distance range ΔL, with respect to the distance Lh to the region center, is used as the score representative value S^n of each recognition model Mn (see below). Figure 18 Additionally, the suffix n in the attached figure S^n refers to an integer from 1 to N, serving as an index for identifying the representative score S^n corresponding to each of the N recognition models Mn.

[0083] Therefore, the fusion block 120 of S105 is defined by Equation 3, where Equation 3 uses the fusion rate ωn of each recognition model Mn to weight the score representative value S^n of each recognition model Mn, and calculates... Figure 18 The weighted average J. Under this definition, as an optimized operation using the fusion rate ωn from Equation 3, the following is adopted. Figure 18 Any one of the operations C1 to C7 shown.

[0084] [Formula 3]

[0085] J=∑ωn·S^n

[0086] Operation C1 optimizes the fusion rate ωn by maximizing the weighted average J(TP+TN) of the sum of the recognition scores Sn associated with the consistency scores TP and TN, respectively. Alternatively, operation C1 optimizes the fusion rate ωn by minimizing the reciprocal of the weighted average J(TP+TN) of the sum of the recognition scores Sn associated with the consistency scores TP and TN, respectively.

[0087] Operation C2 optimizes the fusion rate ωn by maximizing the reciprocal of the weighted average J(FP+FN) of the sum of the identification scores Sn associated with the inconsistencies FP and FN, respectively. Alternatively, operation C2 optimizes the fusion rate ωn by minimizing the weighted average J(FP+FN) of the sum of the identification scores Sn associated with the inconsistencies FP and FN, respectively.

[0088] Operation C3 optimizes the fusion rate ωn in a manner that maximizes the reciprocal of the weighted average J(FP) of the fractional representative values ​​S^n of the identification scores Sn associated with the inconsistency FP. Alternatively, operation C3 optimizes the fusion rate ωn in a manner that minimizes the weighted average J(FP) of the fractional representative values ​​S^n of the identification scores Sn associated with the inconsistency FP.

[0089] Operation C4 optimizes the fusion rate ωn by maximizing the reciprocal of the weighted average J(FN) of the fractional representative values ​​S^n of the identification scores Sn associated with the inconsistency degree FN. Alternatively, operation C4 optimizes the fusion rate ωn by minimizing the weighted average J(FN) of the fractional representative values ​​S^n of the identification scores Sn associated with the inconsistency degree FN.

[0090] Operation C5 optimizes the fusion rate ωn by maximizing the sum of the weighted average J(TP+TN) of the representative values ​​S^n, which are the sums of the identification scores Sn associated with consistency TP and TN respectively, and the reciprocal of the weighted average J(FP+FN) of the representative values ​​S^n, which are the sums of the identification scores Sn associated with inconsistencies FP and FN respectively. Alternatively, operation C5 optimizes the fusion rate ωn by minimizing the sum of the reciprocal of the weighted average J(TP+TN) of the representative values ​​S^n, which are the sums of the identification scores Sn associated with consistency TP and TN respectively, and the weighted average J(FP+FN) of the representative values ​​S^n, which are the sums of the identification scores Sn associated with inconsistencies FP and FN respectively.

[0091] Operation C6 optimizes the fusion rate ωn by maximizing the sum of the weighted average J(TP+TN) of the representative values ​​S^n (which uses the sum of the recognition scores Sn associated with consistency TP and TN respectively) and the reciprocal of the weighted average J(FP) of the representative values ​​S^n (which uses the recognition scores Sn associated with inconsistency FP respectively). Alternatively, operation C6 optimizes the fusion rate ωn by minimizing the sum of the reciprocal of the weighted average J(TP+TN) of the representative values ​​S^n (which uses the sum of the recognition scores Sn associated with consistency TP and TN respectively) and the weighted average J(FP) of the representative values ​​S^n (which uses the recognition scores Sn associated with inconsistency FP respectively).

[0092] Operation C7 optimizes the fusion rate ωn by maximizing the sum of the weighted average J(TP+TN) of the representative values ​​S^n (which uses the sum of the recognition scores Sn associated with consistency TP and TN respectively) and the reciprocal of the weighted average J(FN) of the representative values ​​S^n (which uses the recognition scores Sn associated with inconsistency FN respectively). Alternatively, operation C7 optimizes the fusion rate ωn by minimizing the sum of the reciprocal of the weighted average J(TP+TN) of the representative values ​​S^n (which uses the sum of the recognition scores Sn associated with consistency TP and TN respectively) and the weighted average J(FN) of the representative values ​​S^n (which uses the recognition scores Sn associated with inconsistency FN respectively).

[0093] The fusion block 120 of S105 weights the recognition data of each recognition model Mn using the optimized fusion rate ωn, and merges the recognition data of each recognition model Mn into unified data. At this time, for example, if the recognition data of each recognition model Mn includes the position coordinates of the target moving body 3, the position coordinates of each recognition model Mn can be weighted using the fusion rate ωn, and the weighted average position coordinates can be output as unified data.

[0094] Furthermore, for example, if the recognition data of each recognition model Mn includes recognition confidence for the target moving body 3 and other types of data, the fusion rate ωn of the recognition model Mn whose recognition confidence is reduced to an unacceptable range can be forced to a value of 0. Here, in the latter example, other types of data can also be fused in recognition models Mn other than the recognition model Mn with reduced recognition confidence by normalizing the fusion rate ωn. In this way, other types of data obtained from the recognition model Mn with lower recognition confidence can be removed from the fusion of the integrated data regardless of the optimized fusion rate ωn.

[0095] The fusion block 120 of S105 can also associate at least one of the optimized fusion rate ωn and the fused identification data with a timestamp and store it in the memory 10. The fusion block 120 of S105 can also store the optimized fusion rate ωn and at least one of the fused identification data as timestamp-associated information via the communication system 5 to an external server. The fusion block 120 of S105 can also display the fused identification data obtained through the optimized fusion rate ωn from the information prompting system 6.

[0096] exist Figure 4 In S106 of the processing flow shown, constraint block 130 defines Equation 4. Equation 4 uses the fusion rate ωn of each optimized recognition model Mn to weight each recognition model Mn according to the recognition score Sn of each grid region Am, and calculates... Figure 19The weighted average K. Under this definition, the constraint block 130 of S105 sets constraints on the Operational Design Domain (ODD) for the main vehicle 2 in autonomous driving mode based on the weighted average K. At this time, if it is the object of the constraint and Figure 19 As illustrated by the × symbol, the grid region Am outside the set range is removed from the ODD by using the weighted average K of the identification scores Sn associated with at least one of the consistency scores TP, TN and the inconsistency scores FP, FN.

[0097] [Formula 4]

[0098] K=∑ωn·Sn

[0099] Here, a weighted average K of the identification score Sn associated with at least one of the consistency degrees TP and TN is used as a limiting trigger for the grid region Am if it falls below the allowable range for the consistency degrees TP and TN. On the other hand, a weighted average K of the identification score Sn associated with at least one of the inconsistencies FP and FN is used as a limiting trigger for the grid region Am if it rises above the allowable range for the inconsistencies FP and FN.

[0100] (Effects)

[0101] The effects of the above-described embodiment will now be explained.

[0102] According to this embodiment, a probability distribution Dp indicating the presence of a target moving body 3 at a distance Lh relative to the driving area Ad of the main vehicle 2 is obtained. Therefore, in this embodiment, recognition data from multiple recognition models Mn are fused based on a fusion rate ωn based on a recognition score Sn, wherein the recognition score Sn is related to the recognition rate and the thus obtained probability distribution Dp for each of the multiple recognition models Mn used to identify the target moving body 3. Thus, even if the recognition performance changes according to the requirements of the driving area Ad, appropriate fusion can be achieved by following a fusion rate ωn that matches the recognition score Sn, which is reflected by the probability distribution Dp. Therefore, the recognition accuracy of the target moving body 3 in the main vehicle 2 can be ensured.

[0103] According to this embodiment, the probability distribution Dp is obtained based on the velocity distribution Dv and acceleration distribution Da of the moving body in the driving area Ad. Therefore, even if the recognition performance is required to vary according to the velocity distribution Dv and acceleration distribution Da that affect the probability distribution Dp in the driving area Ad, appropriate fusion can be achieved using a fusion rate ωn that matches the recognition score Sn, which reflects this variation. Thus, the recognition accuracy of the target moving body 3 in the main vehicle 2 can be improved.

[0104] According to this embodiment, the probability distribution Dp within the range of the larger side's movement distance ΔL in the critical gap tc of the main vehicle 2 moving within the driving area Ad and the recognition time ts of the main vehicle 2 in the driving area Ad is obtained. Therefore, even if the required recognition performance changes according to the probability distribution Dp within the range of the safe side's movement distance ΔL, which is limited to the critical gap tc and the recognition time ts, appropriate fusion can be achieved using a fusion rate ωn that matches the recognition score Sn, which reflects this change. Therefore, the recognition accuracy of the target moving body 3 in the main vehicle 2 can be ensured in the shortest possible processing time.

[0105] According to this embodiment, the fusion rate ωn is optimized by using the recognition score Sn of each recognition model Mn, which is a correction of the recognition rate (aspect of consistency between recognition and reality, TP and TN) based on the probability distribution Dp. This optimizes the fusion rate ωn to ensure consistency between recognition and reality after fusion. Therefore, high-precision recognition accuracy of the target moving body 3 in the main vehicle 2 can be ensured.

[0106] According to this embodiment, the fusion rate ωn is optimized by using the recognition score Sn of each recognition model Mn, which is a correction of the recognition rate based on at least one of the inconsistencies between recognition and reality (FP and FN) using the probability distribution Dp. This optimizes the fusion rate ωn, reducing the inconsistency between recognition and reality after fusion. Therefore, the recognition accuracy of the target moving body 3 in the main vehicle 2 can be ensured with high precision.

[0107] According to this embodiment, the fusion rate ωn is optimized based on the recognition score Sn of each recognition model Mn, after the recognition rate is corrected according to the probability of existence in the probability distribution Dp depending on the distance Lh from the main vehicle 2. Therefore, even if the required recognition performance changes, appropriate fusion can be achieved using a fusion rate ωn that matches the recognition score Sn, which reflects this change according to the distance Lh in the driving area Ad. Thus, the recognition accuracy of the target moving body 3 can be ensured throughout the driving area Ad of the main vehicle 2.

[0108] According to this embodiment, the fusion rate ωn is optimized by using the recognition score Sn of each recognition model Mn, which is obtained by correcting the recognition rate of each grid region Am formed by dividing the driving area Ad into multiple regions using the existence probability of the probability distribution Dp. Therefore, the fusion rate ωn can be precisely optimized for the required recognition performance in each grid region Am corresponding to the distance Lh from the main vehicle 2. Thus, the recognition accuracy of the target moving body 3 throughout the driving area Ad of the main vehicle 2 can be ensured with high precision and in a short time.

[0109] According to this embodiment, the ODD (Optical Distribution Decision) for the main vehicle 2 in autonomous driving mode is set based on a weighted average K obtained by weighting the recognition scores Sn of each recognition model Mn and each grid region Am using a fusion rate ωn. Therefore, even if the recognition performance changes according to the requirements of the driving area Ad, the main vehicle 2 in autonomous driving mode can be subject to an ODD that matches the recognition score Sn and fusion rate ωn that reflect this change. Thus, in the main vehicle 2, in addition to the recognition accuracy of the target moving body 3, the accuracy of the ODD setting can also be ensured.

[0110] (Other implementation methods)

[0111] The above describes one embodiment, but this disclosure is not intended to be limited to the described embodiment and can be applied to various embodiments without departing from the spirit of this disclosure.

[0112] In a variation, the dedicated computer constituting the processing system 1 may also have at least one of digital circuitry and analog circuitry as a processor. Here, the so-called digital circuitry includes, for example, at least one of ASIC (Application Specific Integrated Circuit), FPGA (Field Programmable Gate Array), SOC (System on a Chip), PGA (Programmable Gate Array), and CPLD (Complex Programmable Logic Device). Furthermore, such digital circuitry may also have a memory storing a program.

[0113] In S102 of the modified example performed by the distribution acquisition block 100, the movement distance range ΔL can also be limited to the distance the main vehicle 2 moves within the critical gap tc. In S102 of the modified example performed by the distribution acquisition block 100, the movement distance range ΔL can also be limited to the distance the main vehicle 2 moves within the recognition time ts. In S102 of the modified example performed by the distribution acquisition block 100, the movement distance range ΔL can also be limited to a fixed value other than those in these modifications. In S102 of the modified example performed by the distribution acquisition block 100, the probability distribution Dp specific to each driving area Ad can also be obtained from, for example, the external center.

[0114] In S103 of the modified example performed by the recognition rate acquisition block 110, the recognition rate can also be continuously dispersed relative to the distance Lh for each recognition model Mn in at least a portion of the orientation centered on the main vehicle 2 in the driving area Ad. In this case, in S105 performed by the fusion block 120, the recognition score Sn can also be continuously dispersed relative to the distance Lh for each recognition model Mn in at least a portion of the orientation centered on the main vehicle 2 in the driving area Ad.

[0115] In S103 of the modified example performed by the recognition rate acquisition block 110, the recognition rate of each recognition model Mn can also be acquired as a shared value for the entire region of the driving area Ad. In this case, in S105 performed by the fusion block 120, the recognition score Sn of each recognition model Mn is acquired as a shared value for the entire region of the driving area Ad.

[0116] In S105 of the modified example performed by fusion block 120, the fusion rate ωn can also be optimized by combining multiple operations from C1 to C7. In the modified example, S106 performed by restriction block 130 can also be omitted. In the modified example, the main moving body of the application processing system 1 can be, for example, an autonomous driving robot capable of remote control of its movement.

[0117] In addition to the above description, the above-described embodiments and modifications can also be implemented as a processing device that can be mounted on the main vehicle 2, wherein the processor 12 and memory 10 of the processing system 1 each have at least one control device (e.g., a control ECU). Furthermore, the above-described embodiments and modifications can also be implemented as a processing device that can be mounted on the main vehicle 2, wherein the processor 12 and memory 10 of the processing system 1 each have at least one semiconductor device (e.g., a semiconductor wafer).

Claims

1. A processing system having a processor that performs identification association processing associated with the identification of a target moving body in a main moving body, wherein, The processor described above is configured to perform the following processes: Based on the driving area of ​​the main mobile vehicle, obtain the probability distribution of the existence of the target mobile vehicle relative to the distance from the main mobile vehicle; For each of the multiple recognition models, obtain the recognition rate for the aforementioned moving target. as well as The recognition data of each of the above recognition models are fused according to the fusion rate based on the recognition score, wherein the recognition score is related to the recognition rate and the probability distribution of each of the above recognition models. Obtaining the above probability distribution includes: The probability distribution is obtained based on the velocity and acceleration distributions of the moving objects in the aforementioned driving area. Obtaining the above probability distribution includes: The probability distribution is obtained within the range of the larger side movement distance in the critical gap between the main moving body and the moving body in the driving area and the time required for identification of the main moving body in the driving area.

2. The processing system according to claim 1, wherein, The fusion of the above identification data includes: The fusion rate is optimized based on the recognition score of each of the aforementioned recognition models, wherein the recognition score of each of the aforementioned recognition models is a recognition score after correcting the recognition rate as the consistency of recognition with reality using the aforementioned probability distribution.

3. The processing system according to claim 1, wherein, The fusion of the above identification data includes: The fusion rate is optimized based on the recognition score of each of the aforementioned recognition models, wherein the recognition score of each of the aforementioned recognition models is a recognition score after correcting the recognition rate as the degree of inconsistency between recognition and reality using the aforementioned probability distribution.

4. The processing system according to claim 1, wherein, The aforementioned processing system has a storage medium. The above-mentioned processor configuration is also used to execute: The aforementioned fusion rate is stored in the aforementioned storage medium.

5. A processing system having a processor that performs identification association processing associated with the identification of a target moving body in a main moving body, wherein, The processor described above is configured to perform the following processes: Based on the driving area of ​​the main mobile vehicle, obtain the probability distribution of the existence of the target mobile vehicle relative to the distance from the main mobile vehicle; For each of the multiple recognition models, obtain the recognition rate for recognizing the aforementioned moving target. as well as The recognition data of each of the above recognition models are fused according to the fusion rate based on the recognition score, wherein the recognition score is related to the recognition rate and the probability distribution of each of the above recognition models. Obtaining the above probability distribution includes: The probability distribution is obtained based on the velocity and acceleration distributions of the moving objects in the aforementioned driving area. The fusion of the above identification data includes: The fusion rate is optimized based on the recognition score of each of the above recognition models, wherein the recognition score of each of the above recognition models is a recognition score after correcting the recognition rate using the existence probability of the distance in the above probability distribution.

6. The processing system according to claim 5, wherein, The fusion of the above identification data includes: The fusion rate is optimized based on the recognition score of each of the above recognition models, wherein the recognition score of each of the above recognition models is the recognition score after correcting the recognition rate of each of the grid regions formed by dividing the driving area into multiple regions using the existence probability.

7. The processing system according to claim 6, wherein, The above-mentioned processor configuration is also used to execute: The operating design area for the aforementioned main mobile body in autonomous driving mode is limited by a weighted average, wherein the weighted average is obtained by weighting the recognition scores of each of the aforementioned recognition models and each of the aforementioned grid regions using the aforementioned fusion rate.

8. The processing system according to any one of claims 5 to 7, wherein, The fusion of the above identification data includes: The fusion rate is optimized based on the recognition score of each of the aforementioned recognition models, wherein the recognition score of each of the aforementioned recognition models is a recognition score after correcting the recognition rate as the consistency of recognition with reality using the aforementioned probability distribution.

9. The processing system according to any one of claims 5 to 7, wherein, The fusion of the above identification data includes: The fusion rate is optimized based on the recognition score of each of the aforementioned recognition models, wherein the recognition score of each of the aforementioned recognition models is a recognition score after correcting the recognition rate as the degree of inconsistency between recognition and reality using the aforementioned probability distribution.

10. The processing system according to any one of claims 5 to 7, wherein, The aforementioned processing system has a storage medium. The above-mentioned processor configuration is also used to execute: The aforementioned fusion rate is stored in the aforementioned storage medium.

11. A processing system having a processor that performs identification association processing associated with the identification of a target moving body in a main moving body, wherein, The processor described above is configured to perform the following processes: Based on the driving area of ​​the main mobile vehicle, obtain the probability distribution of the existence of the target mobile vehicle relative to the distance from the main mobile vehicle; For each of the multiple recognition models, obtain the recognition rate for recognizing the aforementioned moving target. as well as The recognition data of each of the above recognition models are fused according to the fusion rate based on the recognition score, wherein the recognition score is related to the recognition rate and the probability distribution of each of the above recognition models. The fusion of the above identification data includes: The fusion rate is optimized based on the recognition score of each of the above recognition models, wherein the recognition score of each of the above recognition models is a recognition score after correcting the recognition rate using the existence probability of the distance in the above probability distribution.

12. A processing system having a processor that performs identification association processing associated with the identification of a target moving body in a main moving body, wherein, The processor described above is configured to perform the following processes: Based on the driving area of ​​the main mobile vehicle, obtain the probability distribution of the existence of the target mobile vehicle relative to the distance from the main mobile vehicle; For each of the multiple recognition models that identify the aforementioned moving target, a recognition rate is obtained, wherein the recognition rate is at least one of the consistency and inconsistency of the recognition relative to reality; as well as The recognition data of each of the above recognition models are fused based on the fusion rate of the recognition score, wherein the recognition score is related to the recognition rate and probability distribution of each of the above recognition models. Obtaining the above probability distribution includes: The probability distribution is obtained based on the velocity and acceleration distributions of the moving objects in the aforementioned driving area. Obtaining the above probability distribution includes: The probability distribution is obtained within the range of the larger side movement distance in the critical gap between the main moving body and the moving body in the driving area and the time required for identification of the main moving body in the driving area.

13. A processing system having a processor that performs identification association processing associated with the identification of a target moving body in a main moving body, wherein, The processor described above is configured to perform the following processes: Based on the driving area of ​​the main mobile vehicle, obtain the probability distribution of the existence of the target mobile vehicle relative to the distance from the main mobile vehicle; For each of the multiple recognition models that identify the aforementioned moving target, a recognition rate is obtained, wherein the recognition rate is at least one of the consistency and inconsistency of the recognition relative to reality; as well as The recognition data of each of the above recognition models are fused based on the fusion rate of the recognition score, wherein the recognition score is related to the recognition rate and probability distribution of each of the above recognition models. The fusion of the above identification data includes: The fusion rate is optimized based on the recognition score of each of the above recognition models, wherein the recognition score of each of the above recognition models is a recognition score after correcting the recognition rate using the existence probability of the distance in the above probability distribution.

14. A processing apparatus having a processor that performs identification association processing associated with the identification of a target moving body in a main moving body, wherein, The processor described above is configured to perform the following processes: Based on the driving area of ​​the main mobile vehicle, obtain the probability distribution of the existence of the target mobile vehicle relative to the distance from the main mobile vehicle; For each of the multiple recognition models, obtain the recognition rate for recognizing the aforementioned moving target. as well as The recognition data of each of the above recognition models are fused according to the fusion rate based on the recognition score, wherein the recognition score is related to the recognition rate and the probability distribution of each of the above recognition models. Obtaining the above probability distribution includes: The probability distribution is obtained based on the velocity and acceleration distributions of the moving objects in the aforementioned driving area. Obtaining the above probability distribution includes: The probability distribution is obtained within the range of the larger side movement distance in the critical gap between the main moving body and the moving body in the driving area and the time required for identification of the main moving body in the driving area.

15. A processing apparatus having a processor that performs identification association processing associated with the identification of a target moving body in a main moving body, wherein, The processor described above is configured to perform the following processes: Based on the driving area of ​​the main mobile vehicle, obtain the probability distribution of the existence of the target mobile vehicle relative to the distance from the main mobile vehicle; For each of the multiple recognition models, obtain the recognition rate for recognizing the aforementioned moving target. as well as The recognition data of each of the above recognition models are fused according to the fusion rate based on the recognition score, wherein the recognition score is related to the recognition rate and the probability distribution of each of the above recognition models. The fusion of the above identification data includes: The fusion rate is optimized based on the recognition score of each of the above recognition models, wherein the recognition score of each of the above recognition models is a recognition score after correcting the recognition rate using the existence probability of the distance in the above probability distribution.

16. A processing apparatus having a processor that performs identification association processing associated with the identification of a target moving body in a main moving body, wherein, The processor described above is configured to perform the following processes: Based on the driving area of ​​the main mobile vehicle, obtain the probability distribution of the existence of the target mobile vehicle relative to the distance from the main mobile vehicle; For each of the multiple recognition models that identify the aforementioned moving target, a recognition rate is obtained, wherein the recognition rate is at least one of the consistency and inconsistency of the recognition relative to reality; as well as The recognition data of each of the above recognition models are fused according to the fusion rate based on the recognition score, wherein the recognition score is related to the recognition rate and probability distribution of each of the above recognition models. Obtaining the above probability distribution includes: The probability distribution is obtained based on the velocity and acceleration distributions of the moving objects in the aforementioned driving area. Obtaining the above probability distribution includes: The probability distribution is obtained within the range of the larger side movement distance in the critical gap between the main moving body and the moving body in the driving area and the time required for identification of the main moving body in the driving area.

17. A processing apparatus having a processor that performs identification association processing associated with the identification of a target moving body in a main moving body, wherein, The processor described above is configured to perform the following processes: Based on the driving area of ​​the main mobile vehicle, obtain the probability distribution of the existence of the target mobile vehicle relative to the distance from the main mobile vehicle; For each of the multiple recognition models that identify the aforementioned moving target, a recognition rate is obtained, wherein the recognition rate is at least one of the consistency and inconsistency of the recognition relative to reality; as well as The recognition data of each of the above recognition models are fused according to the fusion rate based on the recognition score, wherein the recognition score is related to the recognition rate and probability distribution of each of the above recognition models. The fusion of the above identification data includes: The fusion rate is optimized based on the recognition score of each of the above recognition models, wherein the recognition score of each of the above recognition models is a recognition score after correcting the recognition rate using the existence probability of the distance in the above probability distribution.

18. A processing method that is executed by a processor in order to execute identification association processing associated with identification of a target mobile body in a host mobile body, wherein, Include: Based on the driving area of ​​the main mobile vehicle, obtain the probability distribution of the existence of the target mobile vehicle relative to the distance from the main mobile vehicle; For each of the multiple recognition models, obtain the recognition rate for recognizing the aforementioned moving target. as well as The recognition data of each of the above recognition models are fused according to the fusion rate based on the recognition score, wherein the recognition score is related to the recognition rate and the probability distribution of each of the above recognition models. Obtaining the above probability distribution includes: The probability distribution is obtained based on the velocity and acceleration distributions of the moving objects in the aforementioned driving area. Obtaining the above probability distribution includes: The probability distribution is obtained within the range of the larger side movement distance in the critical gap between the main moving body and the moving body in the driving area and the time required for identification of the main moving body in the driving area.

19. A processing method, which is a processing method executed by a processor in order to execute identification-related processing associated with identification of a target mobile body among main mobile bodies, in which, Include: Based on the driving area of ​​the main mobile vehicle, obtain the probability distribution of the existence of the target mobile vehicle relative to the distance from the main mobile vehicle; For each of the multiple recognition models, obtain the recognition rate for recognizing the aforementioned moving target. as well as The recognition data of each of the above recognition models are fused according to the fusion rate based on the recognition score, wherein the recognition score is related to the recognition rate and the probability distribution of each of the above recognition models. The fusion of the above identification data includes: The fusion rate is optimized based on the recognition score of each of the above recognition models, wherein the recognition score of each of the above recognition models is a recognition score after correcting the recognition rate using the existence probability of the distance in the above probability distribution.

20. A processing method that is executed by a processor in order to execute identification association processing associated with identification of a target mobile body in a host mobile body, wherein, Include: Based on the driving area of ​​the main mobile vehicle, obtain the probability distribution of the existence of the target mobile vehicle relative to the distance from the main mobile vehicle; For each of the multiple recognition models that identify the aforementioned moving target, a recognition rate is obtained, wherein the recognition rate is at least one of the consistency and inconsistency of the recognition relative to reality; as well as The recognition data of each of the above recognition models are fused according to the fusion rate based on the recognition score, wherein the recognition score is related to the recognition rate and probability distribution of each of the above recognition models. Obtaining the above probability distribution includes: The probability distribution is obtained based on the velocity and acceleration distributions of the moving objects in the aforementioned driving area. Obtaining the above probability distribution includes: The probability distribution is obtained within the range of the larger side movement distance in the critical gap between the main moving body and the moving body in the driving area and the time required for identification of the main moving body in the driving area.

21. A processing method that is executed by a processor in order to execute identification association processing associated with identification of a target mobile body in a host mobile body, wherein, Include: Based on the driving area of ​​the main mobile vehicle, obtain the probability distribution of the existence of the target mobile vehicle relative to the distance from the main mobile vehicle; For each of the multiple recognition models that identify the aforementioned moving target, a recognition rate is obtained, wherein the recognition rate is at least one of the consistency and inconsistency of the recognition relative to reality; as well as The recognition data of each of the above recognition models are fused according to the fusion rate based on the recognition score, wherein the recognition score is related to the recognition rate and probability distribution of each of the above recognition models. The fusion of the above identification data includes: The fusion rate is optimized based on the recognition score of each of the above recognition models, wherein the recognition score of each of the above recognition models is a recognition score after correcting the recognition rate using the existence probability of the distance in the above probability distribution.

22. A storage medium storing a processing program, the processing program being stored in the storage medium for performing identification association processing associated with the identification of a target mobile body in a main mobile body, and the processing program comprising instructions for execution by a processor, wherein, The above instructions include: Based on the driving area of ​​the main mobile vehicle, obtain the probability distribution of the existence of the target mobile vehicle relative to the distance from the main mobile vehicle; For each of the multiple recognition models, obtain the recognition rate for recognizing the aforementioned moving target. as well as The recognition data of each of the above recognition models are fused according to the fusion rate based on the recognition score, wherein the recognition score is related to the recognition rate and the probability distribution of each of the above recognition models. Obtaining the above probability distribution includes: The probability distribution is obtained based on the velocity and acceleration distributions of the moving objects in the aforementioned driving area. Obtaining the above probability distribution includes: The probability distribution is obtained within the range of the larger side movement distance in the critical gap between the main moving body and the moving body in the driving area and the time required for identification of the main moving body in the driving area.

23. A storage medium storing a processing program, the processing program being stored in the storage medium for performing identification association processing associated with the identification of a target mobile body in a main mobile body, and the processing program comprising instructions for causing a processor to execute, wherein, The above instructions include: Based on the driving area of ​​the main mobile vehicle, obtain the probability distribution of the existence of the target mobile vehicle relative to the distance from the main mobile vehicle; For each of the multiple recognition models, obtain the recognition rate for recognizing the aforementioned moving target. as well as The recognition data of each of the above recognition models are fused according to the fusion rate based on the recognition score, wherein the recognition score is related to the recognition rate and the probability distribution of each of the above recognition models. The fusion of the above identification data includes: The fusion rate is optimized based on the recognition score of each of the above recognition models, wherein the recognition score of each of the above recognition models is a recognition score after correcting the recognition rate using the existence probability of the distance in the above probability distribution.

24. A storage medium storing a processing program, the processing program being stored in the storage medium for performing identification association processing associated with the identification of a target mobile body in a main mobile body, and the processing program comprising instructions for causing a processor to execute, wherein, The above instructions include: Based on the driving area of ​​the main mobile vehicle, obtain the probability distribution of the existence of the target mobile vehicle relative to the distance from the main mobile vehicle; For each of the multiple recognition models that identify the aforementioned moving target, a recognition rate is obtained, wherein the recognition rate is at least one of the consistency and inconsistency of the recognition relative to reality; as well as The recognition data of each of the above recognition models are fused according to the fusion rate based on the recognition score, wherein the recognition score is related to the recognition rate and probability distribution of each of the above recognition models. Obtaining the above probability distribution includes: The probability distribution is obtained based on the velocity and acceleration distributions of the moving objects in the aforementioned driving area. Obtaining the above probability distribution includes: The probability distribution is obtained within the range of the larger side movement distance in the critical gap between the main moving body and the moving body in the driving area and the time required for identification of the main moving body in the driving area.

25. A storage medium storing a processing program, the processing program being stored in the storage medium for performing identification association processing associated with the identification of a target mobile body in a main mobile body, and the processing program comprising instructions for causing a processor to execute, wherein, The above instructions include: Based on the driving area of ​​the main mobile vehicle, obtain the probability distribution of the existence of the target mobile vehicle relative to the distance from the main mobile vehicle; For each of the multiple recognition models that identify the aforementioned moving target, a recognition rate is obtained, wherein the recognition rate is at least one of the consistency and inconsistency of the recognition relative to reality; as well as The recognition data of each of the above recognition models are fused according to the fusion rate based on the recognition score, wherein the recognition score is related to the recognition rate and probability distribution of each of the above recognition models. The fusion of the above identification data includes: The fusion rate is optimized based on the recognition score of each of the above recognition models, wherein the recognition score of each of the above recognition models is a recognition score after correcting the recognition rate using the existence probability of the distance in the above probability distribution.

Images