Target tracking device, target tracking method, and program

Abstract

A target tracking device (1) comprises: a sensor information acquisition unit (11) that acquires sensor information; a sensor information discrimination unit (12) that discriminates sensor information for each information type; a tracking filter unit (13) that calculates a smoothed value and a predicted value of a state vector for each peripheral vehicle by a tracking filter process using the sensor information discriminated as information of a peripheral vehicle; a road curvature acquisition unit (14) that acquires a road curvature of a travel road of the vehicle; a predicted route calculation unit (15) that calculates a predicted route of the peripheral vehicle on the basis of the smoothed value of the state vector for each peripheral vehicle and the road curvature of the travel road; and a determination unit (16) that determines the presence or absence of cut-in of the peripheral vehicle to the travel lane of the vehicle on the basis of the residual between the predicted route and the sensor information of the peripheral vehicle at the predicted time.

Description

Target tracking device, target tracking method and program

[0001] The present disclosure relates to a target tracking device, a target tracking method, and a program.

[0002] For the purpose of assisting vehicle driving, a technology for determining whether a nearby vehicle will cut into the vehicle's driving lane has been proposed. For example, Non-Patent Document 1 describes a technology for predicting the behavior of nearby vehicles by solving a route planning problem for searching for a route that allows the nearby vehicle to avoid an obstacle ahead. This technology searches for route candidates that can avoid an obstacle ahead, the position of which is known, and among the multiple route candidates found, the route candidate that maximizes a constraint function is determined as the optimal route.

[0003] Z. Jian, “Obstacle-Centered Trajectory Planning for Autonomous Mobile Robot,” 2021 IEEE Intelligent Transportation Systems Conference (ITSC) Indianapolis, USA. September 19-21, 2021.

[0004] The conventional technology described in Non-Patent Document 1 has a problem in that it is necessary to establish an empirical rule regarding the behavior of surrounding vehicles in order to predict the behavior of surrounding vehicles.

[0005] The present disclosure is intended to solve the above-mentioned problems, and aims to provide a target tracking device that can determine whether a target nearby vehicle has cut in without establishing an empirical rule regarding the behavior of nearby vehicles.

[0006] The target tracking device according to the present disclosure includes a sensor information acquisition unit that acquires sensor information obtained by sensing the area around the vehicle, a sensor information discrimination unit that discriminates the sensor information into information types, a tracking filter unit that calculates smoothed values ​​and predicted values ​​of the state vector for each surrounding vehicle by tracking filter processing using the sensor information discriminated as information about the surrounding vehicles, a road curvature acquisition unit that acquires the road curvature of the road on which the vehicle is traveling, a predicted route calculation unit that calculates the predicted route of the surrounding vehicles based on the smoothed value of the state vector for each surrounding vehicle and the road curvature of the road on which the vehicle is traveling, and a determination unit that determines whether or not a surrounding vehicle will cut into the vehicle's traveling lane based on the residual between the predicted route and the sensor information of the surrounding vehicle at the predicted time.

[0007] According to the present disclosure, a tracking filter process is performed using sensor information from the surrounding vehicles to calculate smoothed values ​​and predicted values ​​of the state vectors for each surrounding vehicle, a predicted path of the surrounding vehicles is calculated based on the smoothed value of the state vector for each surrounding vehicle and the road curvature of the traveling road, and whether or not a surrounding vehicle will cut in to the traveling lane of the vehicle is determined based on the residual between the predicted path and the sensor information of the surrounding vehicle at the predicted time. In this way, the target tracking device according to the present disclosure can determine whether a surrounding vehicle, which is a target, will cut in to the traveling lane of the vehicle, without using empirical rules regarding the behavior of surrounding vehicles.

[0008] FIG. 4A is a block diagram showing an example of the configuration of a target tracking device according to embodiment 1. FIG. 4B is a flowchart showing a target tracking method according to embodiment 1. FIG. 4C is a diagram showing an overview of a process for calculating the road curvature of a travel road based on sensor information of an object present on the side of the travel road. FIG. 4A and FIG. 4B are block diagrams showing a hardware configuration for realizing the functions of the target tracking device according to embodiment 1. FIG. 4C is a block diagram showing an example of the configuration of a target tracking device according to embodiment 2. FIG. 4D is a flowchart showing a target tracking method according to embodiment 2. FIG. 4E is a diagram showing the acceleration of a nearby vehicle cutting into a travel lane.

[0009] Embodiment 1 In embodiment 1, a target tracking device will be described that calculates smoothed values ​​and predicted values ​​of a state vector for each surrounding vehicle by tracking filter processing using sensor information from the surrounding vehicles, calculates a predicted path for each surrounding vehicle based on the smoothed value of the state vector for each surrounding vehicle and the road curvature of the road on which the vehicle is traveling, and determines whether or not a surrounding vehicle will cut into the lane on which the vehicle is traveling based on the residual between the calculated predicted path and the sensor information of the surrounding vehicle at the predicted time.

[0010] (Basic Configuration of Target Tracking Device) FIG. 1 is a block diagram showing an example configuration of a target tracking device 1 according to a first embodiment. In FIG. 1, the target tracking device 1 is mounted on a vehicle and is a target tracking device in which a target to be tracked is a nearby vehicle. Note that the target tracking device 1 is assumed to be a device mounted on a vehicle, but is not limited to this. For example, the target tracking device 1 may be an external device capable of wireless communication with an in-vehicle communication device. The in-vehicle communication device acquires sensor information from an in-vehicle sensor mounted on the vehicle and a roadside sensor installed on the roadside, and transmits the acquired sensor information to the target tracking device 1 via wireless communication. The target tracking device 1 senses the surroundings of the vehicle using the sensor information from the in-vehicle communication device. The determination result is transmitted from the target tracking device 1 to the in-vehicle communication device. The in-vehicle communication device may output the determination result received from the target tracking device 1 to an in-vehicle monitor or the like.

[0011] An information processing application for providing the target tracking service may be installed in the target tracking device 1, or the information processing service may be provided in the form of what is called SaaS (Software as a Service). When the information processing service is provided in the form of SaaS, the information processing application for providing the target tracking service is executed by a server that can be connected to by communication from the target tracking device 1, so that the target tracking service is provided on a web browser without an information processing app being installed in the target tracking device 1.

[0012] 1 , the target tracking device 1 includes a sensor information acquisition unit 11, a sensor information discriminator 12, a tracking filter unit 13, a road curvature acquisition unit 14, a predicted route calculation unit 15, and a determination unit 16. For example, the target tracking device 1 is realized by a computer. A memory included in the computer stores programs constituting an information processing application for realizing the functions of the sensor information acquisition unit 11, the sensor information discriminator 12, the tracking filter unit 13, the road curvature acquisition unit 14, the predicted route calculation unit 15, and the determination unit 16. A processor included in the computer reads the information processing application from the memory and executes the information processing application, thereby realizing the functions of the sensor information acquisition unit 11, the sensor information discriminator 12, the tracking filter unit 13, the road curvature acquisition unit 14, the predicted route calculation unit 15, and the determination unit 16.

[0013] (Sensor Information Acquisition Unit) The sensor information acquisition unit 11 acquires sensor information obtained by sensing the surroundings of the vehicle. For example, the sensor information acquisition unit 11 uses a communication device mounted on the vehicle to establish a communication connection with an on-board sensor and acquire sensor information that detects the state of the road on which the vehicle is traveling, as well as surrounding vehicles present around the vehicle. The on-board sensor is a sensor that is mounted on the vehicle and can detect the surroundings of the vehicle, and examples thereof include millimeter-wave radar, an external camera, and LiDAR.

[0014] The sensor information may include, for example, detection information of surrounding vehicles and detection information of road conditions around the vehicle, and may also include information showing the results of tracking targets such as surrounding vehicles for each sensor using information detected in time series.

[0015] Furthermore, the sensor information may be the result of integrating detection information from each sensor through sensor fusion processing, which is processing for integrating detection information from sensors with different characteristics, such as millimeter-wave radar, LiDAR, or a camera.

[0016] "Integration" may involve directly integrating raw sensor information. For example, a three-dimensional model may be generated by combining three-dimensional point cloud data from LiDAR with images captured by a camera. Features extracted from sensor information may also be combined. For example, sensor information may be information obtained by extracting features related to surrounding vehicles from millimeter-wave radar sensor information and camera images, and combining these features. Furthermore, the results of analyzing the sensor information from each sensor may be combined.

[0017] (Sensor Information Discriminator) The sensor information discriminator 12 discriminates sensor information by information type. For example, the sensor information discriminator 12 acquires raw sensor information detected by various sensors from the sensor information acquirer 11. The sensor information discriminator 12 performs preprocessing such as noise removal, data interpolation, and data format standardization according to the characteristics of each sensor. Note that the sensor information preprocessing may be performed by the sensor information acquirer 11.

[0018] The sensor information discriminator 12 performs sensor fusion processing on the sensor information detected by the multiple sensors. As described above, this processing may be performed by the sensor information acquirer 11. For example, a detailed 3D model of the vehicle's surroundings is constructed by combining images captured by a camera with point cloud data from a LiDAR.

[0019] The sensor information discriminator 12 extracts features from the sensor information integrated by the sensor fusion process and classifies the extracted features by information type. For example, the sensor information discriminator 12 classifies objects such as surrounding vehicles and road conditions using a deep learning or clustering method. Next, the sensor information discriminator 12 discriminates the sensor information by information type based on the classified object information. Information types include surrounding vehicles, obstacles, road conditions, pedestrians, obstacles, etc.

[0020] The sensor information discriminator 12 may also discriminate sensor information by information type using data associations handled in multi-target tracking. In multi-target tracking, it is necessary to continue tracking multiple targets simultaneously. Data associations indicate how sensor information obtained from multiple sensors is associated with existing tracking targets.

[0021] The sensor information discriminator 12 predicts the current position of each target to be tracked based on a movement model for that target, compares the target position indicated by the sensor information with the predicted position of each target, and determines which sensor information obtained by which sensor corresponds to which target. For example, the sensor information discriminator 12 may extract the target position indicated by the sensor information and the predicted position of each target that are closest to each other, and associate them. The sensor information discriminator 12 may also probabilistically evaluate the association between the sensor information and the target and select a combination that is highly associated with the sensor information. Alternatively, the sensor information discriminator 12 may track association hypotheses between multiple pieces of sensor information and continue tracking the target based on each hypothesis.

[0022] (Tracking Filter Unit) The tracking filter unit 13 calculates smoothed values ​​and predicted values ​​of the state vector for each surrounding vehicle by performing tracking filter processing using sensor information discriminated as information about the surrounding vehicle. Examples of tracking filters include a Kalman filter. A Kalman filter estimates the time-varying state around the vehicle from the sensor information. When tracking surrounding vehicles, the Kalman filter can estimate state vectors such as the vehicle's position, speed, or acceleration.

[0023] (Road Curvature Acquisition Unit) The road curvature acquisition unit 14 acquires the road curvature of the road on which the vehicle is traveling. For example, the road curvature acquisition unit 14 calculates the road curvature of the road on which the vehicle is traveling based on time-series data of sensor information that detects the state of the road on which the vehicle is traveling or time-series data of the predicted positions of surrounding vehicles at future times, and acquires the calculation result of the road curvature. Specifically, for example, the road curvature acquisition unit 14 extracts features related to the shape of the road from the time-series data of sensor information that detects the state of the road, and constructs a model for calculating the road curvature from the extracted features. For example, the road is modeled as a polynomial function such as a quadratic or cubic function. The road curvature is calculated from the modeled road shape. Here, since the shape of the road may be changing, the road curvature acquisition unit 14 predicts changes in the road curvature based on the direction in which the vehicle is traveling. For example, the road curvature acquisition unit 14 learns change patterns of the road curves or gradients from the time-series data and predicts the road shape at the predicted position of the vehicle at the next time. For this, a prediction model may be constructed using machine learning or state estimation techniques.

[0024] The road curvature acquisition unit 14 may also acquire the road curvature of the traveled road from map data stored in the storage unit 2. For example, the storage unit 2 stores high-precision map data such as a locator. The high-precision map data includes detailed road data such as the positions of road lane markings, traffic signs, traffic lights, and curbs. The road curvature acquisition unit 14 identifies the position of a nearby vehicle traveling ahead from the sensor information, and acquires the curvature of the road along which the vehicle will travel from the map data stored in the storage unit 2 based on the identified position.

[0025] (Predicted Route Calculation Unit) The predicted route calculation unit 15 calculates the predicted routes of the nearby vehicles based on the smoothed value of the state vector of each nearby vehicle and the road curvature of the road. For example, the predicted route calculation unit 15 predicts the positions of the nearby vehicles based on the smoothed value of the state vector of each nearby vehicle and the road curvature of the road, and obtains the predicted routes of the nearby vehicles by continuously connecting the predicted positions.

[0026] (Determination Unit) The determination unit 16 determines whether or not a nearby vehicle will cut in to the vehicle's driving lane based on the residual between the predicted path of the nearby vehicle and the sensor information of the nearby vehicle at the predicted time. For example, the determination unit 16 calculates the residual between the predicted path of the nearby vehicle and the sensor information of the nearby vehicle at the predicted time, and determines whether or not each nearby vehicle will cut in based on the residual accumulated over time. The residual is the difference between the predicted position of the nearby vehicle at the predicted time and the actual position of the nearby vehicle calculated from the sensor information, and indicates how much the nearby vehicle deviates from the predicted path.

[0027] The determination unit 16 determines which lane the surrounding vehicle is currently in based on the residual, and if the residual is within a certain range, it determines that the vehicle will change lanes along the predicted path. The "range" here is set based on, for example, the width of the lane or the lane division of the road. The determination unit 16 checks whether the predicted path of the surrounding vehicle at the predicted time intersects with the vehicle's driving lane, and if so, determines that the surrounding vehicle is cutting in.

[0028] Next, a target tracking method according to the first embodiment will be described. Fig. 2 is a flowchart showing the target tracking method according to the first embodiment, illustrating a series of operations performed by the target tracking device 1. The sensor information acquisition unit 11 acquires sensor information obtained by sensing the surroundings of the vehicle (step ST1). For example, the sensor information acquisition unit 11 uses a communication device mounted on the vehicle to establish a communication connection with an on-board sensor and acquire sensor information that detects the state of the road on which the vehicle is traveling, as well as information about surrounding vehicles present around the vehicle.

[0029] Next, the sensor information discriminator 12 discriminates the sensor information acquired by the sensor information acquirer 11 into sensor information for each information type (step ST2). For example, the sensor information discriminator 12 discriminates the sensor information detected by each sensor from the sensor information acquirer 11 into nearby vehicle information and roadside information.

[0030] The tracking filter unit 13 calculates a smoothed value and a predicted value of the state vector of each surrounding vehicle by performing tracking filter processing using the sensor information of each surrounding vehicle discriminated by the sensor information discriminator 12 (step ST3). For example, the tracking filter unit 13 calculates a smoothed value and a predicted value of the state vector of each surrounding vehicle, which is the prediction target, expressed by the following formula (1): ｋ and the observation vector Z of the surrounding vehicle expressed by the following equation (2): ｋ The state vector X of the surrounding vehicles is obtained. ｋ is a four-dimensional vector whose elements are the positions and velocities of the surrounding vehicles. ｋ is based on the road coordinate system, which is a two-dimensional coordinate system. Road coordinates are coordinates seen from above. x in road coordinates ｋ indicates the longitudinal position of the surrounding vehicle, i.e., the x-coordinate position, and y ｋ indicates the lateral position and y-coordinate position of the surrounding vehicle. ｋ The dots indicate the longitudinal speed of the surrounding vehicle, i.e., the x-coordinate component of the speed, and the y ｋ The dots indicate the lateral speed of the surrounding vehicle, that is, the y-coordinate component of the speed. Note that in the text of the specification, due to the nature of electronic filing, the symbol "・" cannot be placed above the letters, so the x ｋ It is written as a dot. Observation vector Z of surrounding vehicles ｋ is a four-dimensional vector whose elements are the observed positions and observed velocities of the surrounding vehicles in the sensor information of the surrounding vehicles discriminated by the sensor information discriminator 12. ｋ The coordinates are also assumed to be road coordinates.

[0031] The motion model of the target surrounding vehicle is the state vector X ｋ In the following equation (3), Φ ｋ is the state transition matrix. ｋ is the driving noise vector.

[0032] If the target is assumed to move in a uniform linear motion, the state transition matrix Φ ｋcan be expressed by the following equation (4): In the following equation (4), I is a 2×2 unit matrix, 0 is a 2×2 zero matrix, and Δt represents the sampling interval.

[0033] The second term on the right side of the above equation (4), the driving noise vector w ｋ It is assumed that the relationship of the following formula (5) and formula (6) is satisfied. The following formula (5) expresses the driving noise vector w ｋ The following equation (6) shows that the covariance matrix of the driving noise vector is Q ｋ The covariance matrix represents the relationship between each element in the vector and quantifies the behavior and variability of the noise. ｋ describes the properties of the variance and correlation of the driving noise. If these relations are satisfied, the driving noise vector is called a "zero-mean Gauss-Markov process", which has zero mean and the variance of the driving noise at each time step is governed by the covariance matrix Q ｋ This means that it has the property defined by

[0034] The observation model for the observation value of the target to be tracked expressed by the above formula (2) is expressed by the following formula (7). In the following formula (7), H is an observation matrix, and can be expressed by the following formula (8) using the above formulas (1) and (2). That is, H is expressed by a 2 × 2 unit matrix I. v in the following formula (7) ｋ is the sampling time t ｋ Observation vector Z in ｋ is a two-dimensional normally distributed white noise with a mean of 0. Therefore, the observed noise vector v ｋ Similarly to the drive noise, R satisfies the relationships of the following formulas (9) and (10). ｋ is the sampling time t ｋ is the observation noise covariance matrix in the ｋ and the observation noise vector v ｋ are independent of each other.

[0035] The tracking filter unit 13 calculates the predicted value X of the state vector of the surrounding vehicle according to the following equations (11) and (12): ｋ Calculate the underscore (-). Note that in the text of the specification, due to the nature of electronic filing, it is not possible to add an underscore (_) symbol under a letter, and it is not possible to add a hat (^) symbol above a letter. For this reason, X ｋ The following equation (11) is the predicted value X of the state vector at time k. ｋ This is the relational expression for calculating the underscore (-). ｋ The underscore (-) indicates the predicted value before updating the observed value. ｋ－１ represents the state transition from time k-1 to time k. The predicted value X at time k-1 ｋ－１ The underscore hat is the state transition matrix Φ ｋ－１ By performing state transition using ｋ The following equation (12) is the predicted value of the covariance matrix of the state vector at time k, P ｋ This is the relational expression for calculating (-). ｋ－１ (+) is the covariance matrix at time k-1, and Φ ｋ－１ is the state transition matrix. The predicted value P of the covariance matrix shown in the following equation (12) ｋ (-) is the covariance matrix at time k-1, Φ ｋ－１ By using ｋ－１ and the control matrix Γ ｋ－１ This is predicted taking into account the following:

[0036] Furthermore, the tracking filter unit 13 calculates the smoothed value X of the state vector of the surrounding vehicle according to the following equations (13), (14), and (15): ｋ The following equation (13) calculates the smoothing gain K ｋ This is a relational expression for calculating the smoothing gain K ｋis used to adjust how sensitively the filter responds to fluctuations in data. The larger the gain, the more faithfully the filter responds to the original data, but the more likely it is that noise will be included. On the other hand, a small gain reduces noise, but makes the filter less sensitive to sudden changes in data. In the following equation (13), P ｋ (-) is the predicted value of the covariance matrix of the state vector at the next time k. H is the observation matrix, and R ｋ is the covariance matrix of the observation noise. By calculating the inverse matrix of the covariance matrix using the following equation (13), the smoothing gain K ｋ The following equation (14) is used to calculate the smoothed value X of the state vector. ｋ This is the relational expression for calculating the underscore (+). ｋ The underscore is the predicted value of the state at the filtering step, and Z ｋ is the observation vector at time k, and H is the observation matrix. The following equation (15) is the smoothed value P ｋ In the following formula (15), P ｋ (-) is the predicted value of the covariance matrix of the state vector at the next time k. I is the identity matrix. By updating the covariance matrix taking into account the influence of the observation using the following equation (15), it is possible to calculate a smoothed value of the covariance matrix at the next time k.

[0037] Next, the road curvature acquisition unit 14 calculates and acquires the road curvature of the traveled road as the surrounding environment of the vehicle (step ST4). FIG. 3 is a diagram showing an overview of the process of calculating the road curvature of the traveled road based on sensor information of objects present on the side of the traveled road. In FIG. 3, point A is a time-series data point of the position of the surrounding vehicle that is the target to be tracked, and curve B is the path of the surrounding vehicle determined by these data points A. The road curvature acquisition unit 14 calculates the road curvature ρ of the traveled road of the surrounding vehicle according to the following equation (16). In the following equation (16), L is the horizontal distance between the start point and end point of the curve traveled by the surrounding vehicle, and H is the vertical distance between the apex of the curve and the straight portion.

[0038] The predicted route calculation unit 15 calculates the predicted route of the nearby vehicle based on the smoothed value X of the state vector of each nearby vehicle and the road curvature ρ of the travel road (step ST5). For example, the predicted route calculation unit 15 calculates the predicted route of the nearby vehicle based on the smoothed value X of the state vector of the nearby vehicle calculated by the tracking filter unit 13. ｋ Using the underline hat and the road curvature ρ acquired by the road curvature acquisition unit 14, a predicted path X is calculated for each step time Δ according to the following formulas (17), (18), (19), and (20): ｋ＋Δ Calculate the underscore (-) where ω is the angular velocity of the surrounding vehicle and Δ is the time difference. ｋ The underlined hat indicates the smoothed value X of the state vector. ｋ The velocity component of the surrounding vehicle included in the underscore. ｋ The underbar hat is calculated by the longitudinal speed x of the surrounding vehicle according to the following equation (20): ｋ dot and lateral velocity y ｋ It is calculated using dots.

[0039] The predicted route calculation unit 15 may determine whether the road curvature ρ of the traveled road is within a normal range, and may calculate the predicted route of the surrounding vehicle using a constant-velocity model of the surrounding vehicle without using a road curvature ρ determined to be outside the normal range in calculating the predicted route of the surrounding vehicle. For example, the predicted route calculation unit 15 determines whether the road curvature ρ acquired from the road curvature acquisition unit 14 is within a normal curvature range. The normal curvature range is defined based on, for example, the type of traveled road (highway, general road, etc.), local traffic regulations, or past traffic accident data. The predicted route calculation unit 15 searches for information indicating the normal range stored in the memory unit 2 based on the vehicle position included in the sensor information, and uses the search result information for the above determination. If the road curvature ρ is within the normal range, the predicted route calculation unit 15 calculates the predicted route using the road curvature ρ as described above.

[0040] On the other hand, if the road curvature ρ is outside the normal range, the predicted route calculation unit 15 does not use the road curvature ρ in calculating the predicted route, but calculates the predicted route of the surrounding vehicle using a constant velocity model of the surrounding vehicle. The constant velocity model assumes that the surrounding vehicle moves in a straight line at a constant velocity. The predicted route calculation unit 15 calculates the predicted route of the surrounding vehicle by continuously connecting the predicted positions of the surrounding vehicle predicted using the constant velocity model. This allows the target tracking device 1 to accurately predict the predicted route of the surrounding vehicle.

[0041] The determination unit 16 determines whether or not a nearby vehicle will cut into the vehicle's lane based on the residual between the predicted path of the nearby vehicle and the sensor information of the nearby vehicle at the predicted time (step ST6). For example, the determination unit 16 determines whether or not the nearby vehicle will cut into the vehicle's lane based on the observed vector Z ｋ and the predicted value Hx of the state of the surrounding vehicle at the next time k+Δ ｋ＋Δ The residual (Z ｋ -Hx ｋ＋Δ If the accumulation of S satisfies the relationship shown in the following formula (21), it is determined that there is no cutting in by a nearby vehicle. ｋ is the error between the observed value and the predicted value. The determining unit 16 calculates the error S ｋIf is equal to or smaller than the threshold value d, it is determined that no surrounding vehicle has cut into the lane in which the vehicle is traveling.

[0042] On the other hand, the determination unit 16 determines the residual (Z ｋ -Hx ｋ＋Δ If the accumulation of the error S satisfies the relationship shown in the following formula (22), it is determined that there is no cutting in by a nearby vehicle. ｋ If the error S is greater than the threshold value d, it is determined that a nearby vehicle has cut into the lane in which the vehicle is traveling. ｋ For this, the residual covariance matrix calculated by the tracking filter unit 13 may be used, or parameters within an assumed error range may be applied.

[0043] The determination unit 16 determines whether the process shown in Fig. 2 has been completed (step ST7). If it is determined that the process has been completed (step ST7; YES), the determination unit 16 ends the series of processes shown in Fig. 2. If it is determined that the process has not been completed (step ST7; NO), the determination unit 16 returns to the process of step ST1.

[0044] In determining whether the above process is complete, the determination unit 16 may determine whether the sensor information of the surrounding vehicles is abnormal based on the discriminated sensor information and the predicted state vector for each surrounding vehicle. For example, the determination unit 16 calculates the error between the actual observed value and the predicted state vector, compares the calculated error with a preset threshold, and determines that the sensor information is abnormal if the error exceeds the threshold. For example, the threshold may be calculated by statistically processing the sensor information. If the determination unit 16 determines that the sensor information is abnormal, the sensor information acquisition unit 11 returns to the process of step ST1 and acquires new sensor information about the surrounding vehicles. Next, the sensor information discriminator 12 discriminates the newly acquired sensor information by the sensor information acquisition unit 11. This enables the target tracking device 1 to track a target using normal sensor information.

[0045] Next, a description will be given of the hardware configuration that realizes the functions of the target tracking device 1. The functions of the sensor information acquisition unit 11, sensor information discrimination unit 12, tracking filter unit 13, road curvature acquisition unit 14, predicted route calculation unit 15, and determination unit 16 provided in the target tracking device 1 are realized by processing circuits. That is, the target tracking device 1 includes a processing circuit for executing the processes of steps ST1 to ST7 shown in Fig. 2. The processing circuit may be dedicated hardware, or may be a CPU (Central Processing Unit) that executes a program stored in a memory.

[0046] Fig. 4A is a block diagram showing a hardware configuration for realizing the functions of the target tracking device 1. Fig. 4B is a block diagram showing a hardware configuration for executing software for realizing the functions of the target tracking device 1. In Figs. 4A and 4B, a sensor information acquisition unit 11 acquires sensor information from a sensor external to the device via an input interface 100. A determination unit 16 outputs a determination result to the outside of the device via an output interface 101. For example, the determination unit 16 outputs display information to a display device via the output interface 101 that visually indicates whether or not a nearby vehicle has cut in.

[0047] 4A , the processing circuit 102 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a combination of these. The functions of the sensor information acquisition unit 11, the sensor information discrimination unit 12, the tracking filter unit 13, the road curvature acquisition unit 14, the predicted route calculation unit 15, and the determination unit 16 included in the target tracking device 1 may be realized by separate processing circuits, or these functions may be realized together by a single processing circuit.

[0048] 4B , the functions of the sensor information acquisition unit 11, the sensor information discrimination unit 12, the tracking filter unit 13, the road curvature acquisition unit 14, the predicted route calculation unit 15, and the determination unit 16 included in the target tracking device 1 are realized by software, firmware, or a combination of software and firmware. The software or firmware is written as a program and stored in the memory 104.

[0049] The processor 103 reads and executes programs stored in the memory 104 to realize the functions of the sensor information acquisition unit 11, the sensor information discriminator 12, the tracking filter unit 13, the road curvature acquisition unit 14, the predicted route calculation unit 15, and the determination unit 16 included in the target tracking device 1. For example, the target tracking device 1 includes the memory 104 for storing programs that, when executed by the processor 103, result in the processing of steps ST1 to ST7 shown in FIG. 2 being executed. These programs cause the computer to execute the procedures or methods of the processing performed by the sensor information acquisition unit 11, the sensor information discriminator 12, the tracking filter unit 13, the road curvature acquisition unit 14, the predicted route calculation unit 15, and the determination unit 16. The memory 104 may be a computer-readable storage medium that stores programs for causing the computer to function as the sensor information acquisition unit 11, the sensor information discriminator 12, the tracking filter unit 13, the road curvature acquisition unit 14, the predicted route calculation unit 15, and the determination unit 16.

[0050] The memory 104 may be, for example, a non-volatile or volatile semiconductor memory such as a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable Read Only Memory), or an EEPROM (Electrically-EPROM) (registered trademark), a magnetic disk, a flexible disk, an optical disk, a compact disk, a mini disk, or a DVD.

[0051] Some of the functions of the sensor information acquisition unit 11, sensor information discriminator 12, tracking filter unit 13, road curvature acquisition unit 14, predicted route calculation unit 15, and determination unit 16 included in the target tracking device 1 may be realized by dedicated hardware, and other functions may be realized by software or firmware. For example, the functions of the sensor information acquisition unit 11, sensor information discriminator 12, and tracking filter unit 13 may be realized by a processing circuit 102 that is dedicated hardware, and the functions of the road curvature acquisition unit 14, predicted route calculation unit 15, and determination unit 16 may be realized by a processor 103 reading and executing a program stored in memory 104. In this way, the processing circuit can realize the above functions by hardware, software, firmware, or a combination of these.

[0052] As described above, the target tracking device 1 according to the first embodiment comprises a sensor information acquisition unit 11 that acquires sensor information obtained by sensing the periphery of the vehicle, a sensor information discrimination unit 12 that discriminates the sensor information into information types, a tracking filter unit 13 that calculates smoothed values ​​and predicted values ​​of the state vector for each surrounding vehicle by tracking filter processing using the sensor information discriminated as information about the surrounding vehicles, a road curvature acquisition unit 14 that acquires the road curvature of the road on which the vehicle is traveling, a predicted route calculation unit 15 that calculates a predicted route of the surrounding vehicles based on the smoothed value of the state vector for each surrounding vehicle and the road curvature of the road on which the vehicle is traveling, and a determination unit 16 that determines whether or not a surrounding vehicle will cut in on the vehicle's travel lane based on the residual between the predicted route and the sensor information of the surrounding vehicle at the predicted time. By performing tracking filter processing using sensor information of the surrounding vehicles, smoothed values ​​and predicted values ​​of the state vectors of each surrounding vehicle are calculated, and a predicted path of the surrounding vehicles is calculated based on the smoothed values ​​of the state vectors of each surrounding vehicle and the road curvature of the road on which the vehicle is traveling, and whether or not the surrounding vehicles will cut in on the vehicle's traveling lane is determined based on the residual between the predicted path and the sensor information of the surrounding vehicles at the predicted time.Therefore, the target tracking device 1 can determine whether the target surrounding vehicle will cut in on the vehicle's traveling lane without using any empirical rules regarding the behavior of surrounding vehicles.

[0053] In the target tracking device 1 according to the first embodiment, the determination unit 16 determines whether or not there is an abnormality in the sensor information of the surrounding vehicles based on the deviation state between the sensor information discriminated as information about the surrounding vehicles and the predicted value of the state vector for each surrounding vehicle. The sensor information acquisition unit 11 and the sensor information discriminator 12 acquire new sensor information about the surrounding vehicles whose sensor information has been determined to be abnormal. This enables the target tracking device 1 to track the target using sensor information that is normal.

[0054] In the target tracking device 1 according to the first embodiment, the predicted path calculation unit 15 determines whether the road curvature of the traveled road is within a normal range, and does not use the road curvature determined to be out of the normal range in calculating the predicted paths of the surrounding vehicles, but calculates the predicted paths of the surrounding vehicles using a constant velocity model of the surrounding vehicles. This allows the target tracking device 1 to accurately predict the predicted paths of the surrounding vehicles.

[0055] The target tracking method according to the first embodiment includes the steps of: a step ST1 in which a sensor information acquisition unit 11 acquires sensor information obtained by sensing the periphery of the vehicle; a step ST2 in which a sensor information discriminator 12 discriminates the sensor information into information types; a step ST3 in which a tracking filter unit 13 calculates smoothed values ​​and predicted values ​​of state vectors for each of the surrounding vehicles by performing tracking filter processing using the sensor information discriminated as information about the surrounding vehicles; a step ST4 in which a road curvature acquisition unit 14 acquires the road curvature of the road on which the vehicle is traveling; a step ST5 in which a predicted path calculation unit 15 calculates predicted paths of the surrounding vehicles based on the smoothed values ​​of the state vectors for each of the surrounding vehicles and the road curvature of the road; and a step ST6 in which a determination unit 16 determines whether a surrounding vehicle will cut into the vehicle's traveling lane based on a residual between the predicted path and the sensor information of the surrounding vehicle at the predicted time. By having the target tracking device 1 execute the above method, it is possible to determine whether a surrounding vehicle, which is a target, will cut in without using any empirical rules regarding the behavior of the surrounding vehicles.

[0056] By executing the program according to the first embodiment, the computer functions as a sensor information acquisition unit 11 that acquires sensor information obtained by sensing the periphery of the vehicle, a sensor information discriminator 12 that discriminates the sensor information into information types, a tracking filter unit 13 that calculates smoothed values ​​and predicted values ​​of a state vector for each nearby vehicle by tracking filter processing using the sensor information discriminated as information about the nearby vehicle, a road curvature acquisition unit 14 that acquires the road curvature of the road on which the vehicle is traveling, a predicted path calculation unit 15 that calculates a predicted path of the nearby vehicle based on the smoothed values ​​of the state vector for each nearby vehicle and the road curvature of the road, and a determination unit 16 that determines whether a nearby vehicle will cut in to the vehicle's traveling lane based on the residual between the predicted path and the sensor information of the nearby vehicle at the predicted time. This makes it possible to determine whether a nearby vehicle, which is a target, will cut in to the lane on which the vehicle is traveling, without using any empirical rules regarding the behavior of the nearby vehicle.

[0057] Second Embodiment In a second embodiment, a target tracking device that uses multiple models to calculate information indicating the predicted paths of surrounding vehicles at future times will be described.

[0058] FIG. 5 is a block diagram showing an example of the configuration of a target tracking device 1A according to a second embodiment. In FIG. 5, the target tracking device 1A is mounted on a vehicle and is a target tracking device in which a target to be tracked is a nearby vehicle. Note that the target tracking device 1A is assumed to be a device mounted on a vehicle, but is not limited to this. For example, the target tracking device 1A may be an external device capable of wireless communication with an in-vehicle communication device. The in-vehicle communication device acquires sensor information from an in-vehicle sensor mounted on the vehicle and a roadside sensor installed on the roadside, and transmits the acquired sensor information to the target tracking device 1A via wireless communication. The target tracking device 1A senses the surroundings of the vehicle using the sensor information from the in-vehicle communication device. The determination result is transmitted from the target tracking device 1A to the in-vehicle communication device. The in-vehicle communication device may output the determination result received from the target tracking device 1A to an in-vehicle monitor or the like.

[0059] An information processing application for providing the target tracking service may be installed in the target tracking device 1A, or the information processing service may be provided in the form of what is called SaaS (Software as a Service). When the information processing service is provided in the form of SaaS, the information processing application for providing the target tracking service is executed by a server that can be connected to by communication from the target tracking device 1A, so that the target tracking service is provided on a web browser without an information processing app being installed in the target tracking device 1A.

[0060] As shown in FIG. 5 , the target tracking device 1A includes a sensor information acquisition unit 11, a sensor information discriminator 12, a tracking filter unit 13, a road curvature acquisition unit 14, a predicted route calculation unit 15A, and a determination unit 16A. For example, the target tracking device 1 is realized by a computer. A memory included in the computer stores a program constituting an information processing application for realizing the functions of the sensor information acquisition unit 11, the sensor information discriminator 12, the tracking filter unit 13, the road curvature acquisition unit 14, the predicted route calculation unit 15A, and the determination unit 16A. The processor included in the computer reads the information processing application from the memory and executes the information processing application, thereby realizing the functions of the sensor information acquisition unit 11, the sensor information discriminator 12, the tracking filter unit 13, the road curvature acquisition unit 14, the predicted route calculation unit 15A, and the determination unit 16A. Note that in FIG. 5 , components that function similarly to those in the target tracking device 1 shown in FIG. 1 are designated by the same reference numerals, and redundant description will be omitted.

[0061] (Predicted Route Calculation Unit) The predicted route calculation unit 15A receives the road curvature ρ and the smoothed value of the state vector for each nearby vehicle, and uses a multiple model to output information indicating the predicted route of the nearby vehicle to calculate information indicating the predicted route of the nearby vehicle at a future time. A multiple model is a technique that uses multiple prediction models or control models to adapt to different situations or scenarios. For example, the predicted route calculation unit 15A prepares multiple motion models corresponding to road conditions with different road curvatures ρ, selects an appropriate model based on the road curvature, and constructs a multiple model of the selected models. The predicted route calculation unit 15A uses the constructed multiple model to calculate information indicating the predicted route of the nearby vehicle at a future time.

[0062] (Determination Unit) The determination unit 16A determines whether or not each nearby vehicle will cut into the vehicle's driving lane based on the residual between the predicted path of the nearby vehicle and sensor information of the nearby vehicle at the prediction time and the posterior probability of the multiple model. For example, the determination unit 16A calculates the residual between information about the predicted path of the nearby vehicle calculated by the predicted path calculation unit 15A and the observed value of the state of the nearby vehicle at the prediction time, and determines whether or not each nearby vehicle will cut into the driving lane based on the accumulation of the calculated residual over time and the posterior probability of the motion model selected as the multiple model, and outputs the determination result.

[0063] Next, a target tracking method according to embodiment 2 will be described. Fig. 6 is a flowchart showing the target tracking method according to embodiment 2, illustrating a series of operations performed by the target tracking device 1A. In Fig. 6, the processing from step ST1 to step ST4 is the same as the processing from step ST1 to step ST4 in Fig. 2, and therefore, redundant explanations will be omitted.

[0064] When the road curvature ρ and the smoothed value of the state vector of each nearby vehicle are input, the predicted route calculation unit 15A calculates information indicating the predicted route of the nearby vehicle at a future time using a multiple model that outputs information indicating the predicted route of the nearby vehicle (step ST5A). For example, the predicted route calculation unit 15A calculates the smoothed value X ｋUsing the underline hat and the road curvature ρ acquired by the road curvature acquisition unit 14, a predicted route of a nearby vehicle a certain time ahead is calculated for each step time Δ according to the above formulas (17), (18), (19), and (20), and further a predicted route is calculated using a left / right turning model expressed by the following formulas (23), (24), and (25). Here, the left / right turning model is a motion model for simulating cutting in by a nearby vehicle. In the following formula (23), X ｋ＋Δ，ａ is the state vector of the surrounding vehicle at time k+Δ, a, and X ｋ is the state vector of the surrounding vehicles at time k. ｋ＋Δ is a state transition matrix, which represents the transition of states from time k to time k+Δ. ｋ＋Δ is the control input matrix, and the control input vector u ａ It shows how the acceleration u affects the state. ａ is a control input vector, which has u and v components. η is a parameter representing an angle. The following equation (23) expresses the previous state X ｋ and the control input vector u ａ Using ｋ＋Δ，ａ The control input matrix Γ at time k+Δ is calculated. ｋ＋Δ is calculated according to the following equation (24): In the following equation (24), the first matrix represents the change in motion depending on time Δ, and the second matrix represents the change in direction with the angle η. ａ has elements u and v. a is a parameter indicating the state of the surrounding vehicle. For example, the state of the surrounding vehicle at time k+Δ is a, and its state vector is X ｋ＋Δ，ａ It is expressed as:

[0065] 7 is a diagram showing the acceleration of a nearby vehicle 200 that cuts into a driving lane. ａis the acceleration of the nearby vehicle 200 that cuts into the lane, and can be defined as a velocity component v in a direction along the base axis of a coordinate system with the vehicle speed V as the base axis and a velocity component u in a direction perpendicular to the base axis, as shown in FIG.

[0066] The predicted route calculation unit 15A may determine whether the road curvature ρ of the traveled road is within a normal range, and may calculate the predicted route of the surrounding vehicle using a constant-velocity model of the surrounding vehicle without using a road curvature ρ determined to be outside the normal range in calculating the predicted route of the surrounding vehicle. For example, the predicted route calculation unit 15A may determine whether the road curvature ρ acquired from the road curvature acquisition unit 14 is within a normal curvature range. The normal curvature range may be defined based on, for example, the type of traveled road (e.g., expressway, general road), local traffic regulations, or past traffic accident data. The predicted route calculation unit 15A searches for information indicating the normal range stored in the memory unit 2 based on the vehicle position included in the sensor information, and uses the search result information for the above determination. If the road curvature ρ is within the normal range, the predicted route calculation unit 15A calculates the predicted route using the road curvature ρ as described above.

[0067] On the other hand, if the road curvature ρ is outside the normal range, the predicted route calculation unit 15A does not use the road curvature ρ in calculating the predicted route, but instead calculates the predicted route of the surrounding vehicle using a constant velocity model of the surrounding vehicle. The constant velocity model assumes that the surrounding vehicle moves in a straight line at a constant velocity. The predicted route calculation unit 15A calculates the predicted route of the surrounding vehicle by continuously connecting the predicted positions of the surrounding vehicle predicted using the constant velocity model. This allows the target tracking device 1A to accurately predict the predicted route of the surrounding vehicle.

[0068] The determination unit 16A determines whether each nearby vehicle will cut into the driving lane based on the residual between the predicted path of the nearby vehicle and the sensor information of the nearby vehicle at the prediction time and the posterior probability of the multiple models (step ST6A). For example, the determination unit 16A determines whether each nearby vehicle will cut into the driving lane based on the accumulation of the residual between the predicted information of the nearby vehicle calculated by the predicted path calculation unit 15A and the observation information of the nearby vehicle at the prediction time, based on the above equation (21), and outputs the determination result. The method for determining whether or not each nearby vehicle will cut into the driving lane follows the following equations (26) and (27). In this case, the determination unit 16A determines whether the residual of the position or speed of the nearby vehicle, which is the residual between the information about the predicted path of the nearby vehicle calculated by the predicted path calculation unit 15A and the observation value of the state of the nearby vehicle at the prediction time, is equal to or greater than a predetermined threshold d. The judgment unit 16A judges that the lane keeping model is valid if the condition of the following equation (26) is satisfied, and judges that the lane keeping model is inconsistent if the condition of the following equation (27) is satisfied.

[0069] In addition to the lane keeping model, the determination unit 16A also calculates the posterior probability μ ｋ，ａ (+) is calculated according to the following formula (28): In the following formula (28), the posterior probability μ ｋ，ａ (+) is the posterior probability of the state a of the surrounding vehicle at time k. ｋ，ａ is the likelihood of state a at time k. Transition probability p ｋ，ａｂ is the transition probability from state b to state a at time k. ｋ－１，ｂ (+) is the posterior probability of the state b of the surrounding vehicle at time k-1. N is the total number of states. The numerator of the following equation (28) is the weighted sum of the posterior probabilities for the transition from state b to state a at the previous time k-1. The weighting coefficient is the transition probability p ｋ，ａｂ and likelihood v ｋ，ａ The denominator of the following equation (28) is a weighted sum over all possible state transitions and acts as a normalization constant, which normalizes the value of the numerator to 1 according to the definition of probability.

[0070] Likelihood v ｋ，ａ is a normal distribution approximation of the observation vector using a multivariate normal distribution g(), and can be expressed by the following formula (29). ｋ is the observed state of the surrounding vehicle at time k. ｋ，１ is the state vector for state 1 of the surrounding vehicle at time k. H is the observation matrix, which is used to map the state vector to the observation space. The error covariance matrix P ｋ，ａ is the error covariance matrix of the state a of the surrounding vehicle at time k. The covariance matrix R of the observation noise ｋ is a parameter representing the uncertainty of the observed value of the state of the surrounding vehicles at time k.

[0071] Based on the calculation results using the above formulas (26), (27), (28), and (29), the judgment unit 16A judges that the lane judgment is valid, i.e., that a nearby vehicle has cut into the driving lane, when the condition that the posterior probability of another model (such as a left or right turning model) expressed by the above formula (27) or (28) is equal to or greater than a certain value is met.

[0072] The determination unit 16A determines whether the process shown in Fig. 6 has been completed (step ST7). If it is determined that the process has been completed (step ST7; YES), the determination unit 16A terminates the series of processes shown in Fig. 6. If it is determined that the process has not been completed (step ST7; NO), the determination unit 16A returns to the process of step ST1.

[0073] In determining whether the above process is complete, the determination unit 16A may determine whether the sensor information of the surrounding vehicles is abnormal based on the discriminated sensor information and the predicted state vector for each surrounding vehicle. For example, the determination unit 16A calculates the error between the actual observed value and the predicted state vector, compares the calculated error with a preset threshold, and determines that the sensor information is abnormal if the error exceeds the threshold. For example, the threshold may be calculated by statistically processing the sensor information. If the determination unit 16A determines that the sensor information is abnormal, the sensor information acquisition unit 11 returns to the process of step ST1 and acquires new sensor information about the surrounding vehicles. Next, the sensor information discriminator 12 discriminates the newly acquired sensor information by the sensor information acquisition unit 11. This enables the target tracking device 1A to track a target using normal sensor information.

[0074] As described above, in the target tracking device 1A according to the second embodiment, the predicted path calculation unit 15A receives the road curvature and the smoothed value of the state vector for each nearby vehicle, and uses a multiple model that outputs information indicating the predicted path of the nearby vehicle to calculate information indicating the predicted path of the nearby vehicle at a future time. The determination unit 16A determines whether each nearby vehicle will cut into the vehicle's driving lane based on the residual and the posterior probability of the multiple model. This allows the target tracking device 1A to determine whether a nearby vehicle, which is a target, will cut in without using any empirical rules regarding the behavior of nearby vehicles. Furthermore, the target tracking device 1A can accurately predict the predicted path of the nearby vehicle.

[0075] In the target tracking device 1A according to the second embodiment, the determination unit 16A determines whether or not there is an abnormality in the sensor information of the surrounding vehicles based on the deviation state between the sensor information discriminated as information about the surrounding vehicles and the predicted value of the state vector for each surrounding vehicle. The sensor information acquisition unit 11 and the sensor information discriminator 12 acquire new sensor information about the surrounding vehicles whose sensor information has been determined to be abnormal. This enables the target tracking device 1A to track the target using normal sensor information.

[0076] In the target tracking device 1A according to the second embodiment, the predicted path calculation unit 15A determines whether the road curvature of the traveled road is within a normal range, and does not use the road curvature determined to be out of the normal range in calculating the predicted paths of the surrounding vehicles, but calculates the predicted paths of the surrounding vehicles using a constant velocity model of the surrounding vehicles. This allows the target tracking device 1A to accurately predict the predicted paths of the surrounding vehicles.

[0077] In the target tracking method according to the second embodiment, a predicted path calculation unit 15A calculates information indicating the predicted paths of nearby vehicles at a future time using a multiple model that outputs information indicating the predicted paths of nearby vehicles when road curvature and smoothed values ​​of state vectors for each nearby vehicle are input. A determination unit 16A determines whether each nearby vehicle will cut into the vehicle's driving lane based on the residual and the posterior probability of the multiple model. By having the target tracking device 1A execute the above method, it is possible to determine whether a nearby vehicle, which is a target, will cut in without using empirical rules regarding the behavior of nearby vehicles. Furthermore, the target tracking device 1A can accurately predict the predicted paths of nearby vehicles.

[0078] When a computer executes the program according to the second embodiment, the predicted path calculation unit 15A calculates information indicating the predicted paths of the surrounding vehicles at a future time using a multiple model that outputs information indicating the predicted path when the road curvature of the traveled road and the smoothed value of the state vector of each surrounding vehicle are input. The determination unit 16A determines whether each surrounding vehicle will cut into the vehicle's travel lane based on the residual and the posterior probability of the multiple model. This allows the target tracking device 1A to determine whether a target surrounding vehicle will cut in without using empirical rules regarding the behavior of surrounding vehicles. Furthermore, the target tracking device 1A can determine whether a target surrounding vehicle will cut in without using empirical rules regarding the behavior of surrounding vehicles.

[0079] Various aspects of the present disclosure are summarized below as appendices.

[0080] (Supplementary Note 1) A target tracking device comprising: a sensor information acquisition unit that acquires sensor information sensed around a vehicle; a sensor information discrimination unit that discriminates the sensor information into information types; a tracking filter unit that calculates a smoothed value and a predicted value of a state vector for each of the surrounding vehicles by tracking filter processing using the sensor information discriminated as information about the surrounding vehicles; a road curvature acquisition unit that acquires the road curvature of a road on which the vehicle is traveling; a predicted route calculation unit that calculates a predicted route for the surrounding vehicles based on the smoothed value of the state vector for each of the surrounding vehicles and the road curvature of the road on which the vehicle is traveling; and a determination unit that determines whether or not the surrounding vehicles will cut into the lane of travel of the vehicle based on a residual between the predicted route and the sensor information of the surrounding vehicles at a predicted time. (Supplementary Note 2) The target tracking device according to Supplementary Note 1, wherein the predicted route calculation unit calculates information indicating the predicted route at a future time using a multiple model that outputs information indicating the predicted route when a road curvature of the travel road and a smoothed value of a state vector for each of the surrounding vehicles are input, and the determination unit determines whether or not each of the surrounding vehicles will cut into the travel lane of the vehicle based on the residual and the posterior probability of the multiple model. (Supplementary Note 3) The target tracking device according to Supplementary Note 1 or Supplementary Note 2, wherein the determination unit determines whether or not there is an abnormality in the sensor information based on a deviation state between sensor information discriminated as information about the surrounding vehicles and a predicted value of a state vector for each of the surrounding vehicles, and the sensor information acquisition unit and the sensor information discrimination unit acquire new sensor information for each of the surrounding vehicles whose sensor information is determined to be abnormal. (Supplementary Note 4) The target tracking device according to any one of Supplementary Note 1 to Supplementary Note 3, wherein the predicted route calculation unit determines whether or not a road curvature of the travel road is within a normal range, does not use a road curvature determined to be not within the normal range in calculating the predicted route of the surrounding vehicle, and calculates the predicted route of the surrounding vehicle using a constant velocity model of the surrounding vehicle.(Supplementary Note 5) A target tracking method using a target tracking device, comprising: a sensor information acquisition unit acquiring sensor information obtained by sensing the periphery of the vehicle; a sensor information discrimination unit discriminating the sensor information into information types; a tracking filter unit calculating a smoothed value and a predicted value of a state vector for each of the surrounding vehicles by tracking filter processing using the sensor information discriminated as information about the surrounding vehicles; a road curvature acquisition unit acquiring the road curvature of a road on which the vehicle is traveling; a predicted route calculation unit calculating a predicted route for the surrounding vehicles based on the smoothed value of the state vector for each of the surrounding vehicles and the road curvature of the road; and a determination unit determining whether or not the surrounding vehicle will cut into the driving lane of the vehicle based on a residual between the predicted route and the sensor information of the surrounding vehicle at the predicted time. (Supplementary Note 6) The target tracking method according to Supplementary Note 5, characterized in that the predicted path calculation unit, when inputted with the road curvature of the travel road and a smoothed value of the state vector for each of the surrounding vehicles, calculates information indicating the predicted path of the surrounding vehicles at a future time using a multiple model that outputs information indicating the predicted path, and the determination unit determines whether or not each of the surrounding vehicles will cut into the travel lane of the vehicle based on the residual and the posterior probability of the multiple model. (Supplementary Note 7) A program for causing a computer to function as: a sensor information acquisition unit that acquires sensor information sensed around the vehicle; a sensor information discrimination unit that discriminates the sensor information into information types; a tracking filter unit that calculates smoothed values ​​and predicted values ​​of state vectors for each of the surrounding vehicles by tracking filter processing using the sensor information discriminated as information about the surrounding vehicles; a road curvature acquisition unit that acquires the road curvature of the road on which the vehicle is traveling; a predicted route calculation unit that calculates a predicted route for the surrounding vehicles based on the smoothed values ​​of the state vector for each of the surrounding vehicles and the road curvature of the road on which the vehicle is traveling; and a determination unit that determines whether or not the surrounding vehicles will cut into the lane on which the vehicle is traveling based on the residual between the predicted route and the sensor information of the surrounding vehicles at the predicted time.(Supplementary Note 8) The program described in Supplementary Note 7, characterized in that the predicted route calculation unit, when inputting the road curvature of the travel road and a smoothed value of the state vector for each of the surrounding vehicles, calculates information indicating the predicted route of the surrounding vehicles at a future time using a multiple model that outputs information indicating the predicted route, and the determination unit determines whether each of the surrounding vehicles will cut into the travel lane of the vehicle based on the residual and the posterior probability of the multiple model.

[0081] It is possible to combine the embodiments, modify any of the components of the embodiments, or omit any of the components of the embodiments.

[0082] The target tracking device according to the present disclosure can be used, for example, for vehicle driving assistance.

[0083] 1, 1A Target tracking device, 2 Memory unit, 11 Sensor information acquisition unit, 12 Sensor information discrimination unit, 13 Tracking filter unit, 14 Road curvature acquisition unit, 15, 15A Predicted route calculation unit, 16, 16A Determination unit, 100 Input interface, 101 Output interface, 102 Processing circuit, 103 Processor, 104 Memory, 200 Peripheral vehicle.

Claims

1. A sensor information acquisition unit that acquires sensor information sensed from the area around the vehicle, A sensor information discrimination unit that discriminates sensor information according to information type, A tracking filter unit calculates smoothed and predicted values ​​of the state vector for each of the surrounding vehicles by tracking filter processing using sensor information discriminated as information of surrounding vehicles, A road curvature acquisition unit that acquires the road curvature of the road on which the vehicle travels, A predictive path calculation unit calculates the predicted path of the surrounding vehicles based on the smoothed value of the state vector for each surrounding vehicle and the road curvature of the road they travel on, A determination unit determines whether or not a surrounding vehicle cuts into the lane of the vehicle based on the residual difference between the predicted path and the sensor information of the surrounding vehicle at the predicted time, A target tracking device equipped with [a specific feature].

2. The predicted path calculation unit, upon receiving the road curvature of the road being traveled and the smoothed value of the state vector for each surrounding vehicle, uses a multiplex model that outputs information indicating the predicted path to calculate information indicating the predicted path at a future time, The determination unit determines whether or not each of the surrounding vehicles cuts into the vehicle's lane based on the residual and the posterior probability of the multiple models. The target tracking device according to claim 1.

3. The determination unit determines whether or not there is an abnormality in the sensor information based on the discrepancy between the sensor information identified as information of the surrounding vehicles and the predicted value of the state vector for each of the surrounding vehicles. The sensor information acquisition unit and the sensor information discrimination unit acquire new sensor information for the surrounding vehicle for which an abnormality in the sensor information has been determined. A target tracking device according to claim 1 or 2, characterized by the above.

4. The predicted path calculation unit determines whether the curvature of the road is within the normal range, and does not use the road curvature that is determined to be outside the normal range in calculating the predicted path of the surrounding vehicles. Instead, it calculates the predicted path of the surrounding vehicles using a constant velocity model of the surrounding vehicles. A target tracking device according to claim 1 or 2, characterized by the above.

5. A target tracking method using a target tracking device, The sensor information acquisition unit acquires sensor information sensed from the area around the vehicle. The sensor information discrimination unit performs the step of discriminating sensor information according to information type, The tracking filter unit performs tracking filter processing using sensor information that has been identified as information about surrounding vehicles to calculate smoothed and predicted values ​​of the state vector for each of the surrounding vehicles. The road curvature acquisition unit performs the steps of acquiring the road curvature of the road on which the vehicle travels, The predictive path calculation unit calculates the predicted path of the surrounding vehicles based on the smoothed value of the state vector for each surrounding vehicle and the road curvature of the road they travel on. The determination unit determines whether or not the surrounding vehicle cuts into the vehicle's lane based on the residual difference between the predicted path and the sensor information of the surrounding vehicle at the predicted time. A target tracking method comprising the following features.

6. When the predicted path calculation unit receives the road curvature of the road being traveled and the smoothed value of the state vector for each of the surrounding vehicles as input, it uses a multiplex model that outputs information indicating the predicted path to calculate information indicating the predicted path of the surrounding vehicles at a future time, The determination unit determines whether or not each of the surrounding vehicles cuts into the vehicle's lane based on the residual and the posterior probability of the multiple models. The target tracking method according to feature 5.

7. Computers, Sensor information acquisition unit that acquires sensor information sensed from the area around the vehicle. Sensor information discrimination unit that discriminates sensor information according to information type, A tracking filter unit calculates smoothed and predicted values ​​of the state vector for each of the surrounding vehicles by tracking filter processing using sensor information discriminated as information of surrounding vehicles. A road curvature acquisition unit that acquires the road curvature of the road on which the vehicle travels. A predictive path calculation unit calculates the predicted path of the surrounding vehicles based on the smoothed value of the state vector for each surrounding vehicle and the road curvature of the road, and A determination unit determines whether or not a surrounding vehicle cuts into the lane of the vehicle based on the residual difference between the predicted path and the sensor information of the surrounding vehicle at the predicted time. A program designed to function as such.

8. When the predicted path calculation unit receives the road curvature of the road being traveled and the smoothed value of the state vector for each of the surrounding vehicles as input, it uses a multiplex model that outputs information indicating the predicted path to calculate information indicating the predicted path of the surrounding vehicles at a future time, The determination unit determines whether or not each of the surrounding vehicles cuts into the vehicle's lane based on the residual and the posterior probability of the multiple models. The program according to feature 7.

Images