Target tracking method and device, vehicle and storage medium

Abstract

The application relates to the technical field of vehicles, in particular to a target tracking method and device, a vehicle and a storage medium, wherein the method comprises the following steps: obtaining a target around the vehicle which has established tracking; calculating a prior probability of the target under at least one motion model and a predicted value of the target under the at least one motion model; obtaining an observation value of the target under the at least one motion model by associating the predicted value with target observation data at a current moment; and calculating a posterior probability of the target under the at least one motion model based on the observation value, the predicted value and the prior probability. The state information of the target is calculated based on the posterior probability, so that the tracking of the target is completed. Thus, the problem that the prediction accuracy of the state information of the target is low in the related art, and the safety and comfort of driving are affected is solved.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and in particular to a target tracking method, device, vehicle, and storage medium. Background Technology

[0002] Currently, traditional vehicle perception fusion methods (i.e., methods for determining the state information of a target during driving) typically rely on a single linear vehicle kinematics model when processing vehicle dynamic perception. While this method provides sufficient performance in most cases, it has limitations in certain specific scenarios. In particular, when faced with complex maneuvers such as sudden braking, rapid acceleration, quick lane changes, or navigating sharp curves, traditional vehicle perception fusion methods may encounter the following problems, taking a vehicle as an example:

[0003] Slow convergence of speed and acceleration: Due to the use of only a single model, there may be a lag in the estimation of vehicle speed and acceleration, leading to a significant deviation between the prediction results and the actual situation; Poor scene adaptability: The impact of the current driving scene on the subsequent movement of surrounding vehicles is not considered, especially in complex scenarios such as intersections, curves, or multi-vehicle road sections, where traditional methods struggle to accurately predict vehicle behavior; Delayed planning and control timing: Inaccurate estimation of vehicle speed and acceleration may lead to delayed subsequent vehicle planning and control (such as acceleration and deceleration). Therefore, current vehicle fusion perception methods have low accuracy in predicting the state information of targets, thus affecting driving safety and comfort. Summary of the Invention

[0004] This application provides a target tracking method, device, vehicle, and storage medium to solve the problems in related technologies, such as low accuracy in predicting the state information of targets, which in turn affects driving safety and comfort.

[0005] The first aspect of this application provides a target tracking method, including the following steps: acquiring targets around the vehicle that have been tracked; calculating the prior probability of the target under at least one motion model, and calculating the predicted value of the target under at least one motion model; associating the predicted value with the target observation data at the current time to obtain the observation value of the target under at least one motion model, and calculating the posterior probability of the target under at least one motion model based on the observation value, the predicted value, and the prior probability; and calculating the state information of the target based on the posterior probability to complete the tracking of the target.

[0006] Optionally, in one embodiment of this application, the motion model includes a uniform acceleration motion model, a uniform circular motion model, and a uniform linear motion model, and the target observation data includes at least one of visual dynamic target information, millimeter-wave radar target information, vehicle odometer information, and lidar target information.

[0007] Optionally, in one embodiment of this application, before calculating the prior probability of the target under at least one motion model, the method further includes: obtaining the vehicle information, lane information, and relative relationship between the vehicle and the target at the current time; and updating the target's motion model based on at least one of the vehicle information, lane information, and relative relationship between the vehicle and the target at the current time.

[0008] Optionally, in one embodiment of this application, updating the motion model of the target based on at least one of the current vehicle information, lane information, and the relative relationship between the vehicle and the target includes: if the lane information indicates that there are no lane lines, then the motion model of the target is a uniform acceleration motion model; if the lane information indicates that the lane curvature is greater than a preset value, then a uniform circular motion model is added to the motion model of the target; if the vehicle information indicates that the vehicle is in a waiting state at an intersection and the target is on the lane-changing side of the vehicle, then a uniform circular motion model is added to the motion model of the target; if the distance between the target and the vehicle meets a preset condition, then a uniform circular motion model or a uniform linear motion model is added to the motion model of the target.

[0009] Optionally, in one embodiment of this application, obtaining the target's observation value under at least one motion model by associating the predicted value with the target observation data at the current time includes: synthesizing the predicted value into a mixed predicted value based on prior probability; and associating the mixed predicted value with the target observation data at the current time to obtain the target's observation value under at least one motion model.

[0010] Optionally, in one embodiment of this application, the posterior probability of the target under at least one motion model is calculated based on the observed value, the predicted value, and the prior probability, including: calculating the likelihood value between the observed value and the predicted value; and calculating the posterior probability of the target under at least one motion model based on the likelihood value and the prior probability.

[0011] Optionally, in one embodiment of this application, calculating the target's state information based on the posterior probability includes: calculating the mean and covariance of the target's posterior probability under all motion models; calculating the mixed posterior mean and posterior covariance based on the mean and covariance respectively, and using the mixed posterior mean and posterior covariance as the target's state information.

[0012] A second aspect of this application provides a target tracking device, comprising: an acquisition module for acquiring targets already tracked around the vehicle; a first calculation module for calculating the prior probability of the target under at least one motion model and calculating the predicted value of the target under at least one motion model; a second calculation module for associating the predicted value with target observation data at the current time to obtain the observation value of the target under at least one motion model, and calculating the posterior probability of the target under at least one motion model based on the observation value, the predicted value, and the prior probability; and a third calculation module for calculating the state information of the target based on the posterior probability to complete the tracking of the target.

[0013] Optionally, in one embodiment of this application, the motion model includes a uniform acceleration motion model, a uniform circular motion model, and a uniform linear motion model, and the target observation data includes at least one of visual dynamic target information, millimeter-wave radar target information, vehicle odometer information, and lidar target information.

[0014] Optionally, in one embodiment of this application, it further includes: an update module, configured to obtain the vehicle information, lane information, and relative relationship between the vehicle and the target at the current moment before calculating the prior probability of the target under at least one motion model; and update the motion model of the target based on at least one of the vehicle information, lane information, and relative relationship between the vehicle and the target at the current moment.

[0015] Optionally, in one embodiment of this application, the updating module is further configured to: if the lane information indicates that there are no lane lines, then the motion model of the target is a uniform acceleration motion model; if the lane information indicates that the lane curvature is greater than a preset value, then a uniform circular motion model is added to the motion model of the target; if the vehicle information indicates that the vehicle is in a waiting state at an intersection and the target is on the lane-changing side of the vehicle, then a uniform circular motion model is added to the motion model of the target; if the distance between the target and the vehicle meets a preset condition, then a uniform circular motion model or a uniform linear motion model is added to the motion model of the target.

[0016] Optionally, in one embodiment of this application, the second calculation module is further configured to: synthesize a mixed prediction value based on prior probabilities; and correlate the mixed prediction value with the target observation data at the current time to obtain the target observation value under at least one motion model.

[0017] Optionally, in one embodiment of this application, the second calculation module is further configured to: calculate the likelihood value between the observed value and the predicted value; and calculate the posterior probability of the target under at least one motion model based on the likelihood value and the prior probability.

[0018] Optionally, in one embodiment of this application, the third calculation module is further configured to: calculate the mean and covariance of the posterior probability of the target under all motion models; calculate the mixed posterior mean and posterior covariance based on the mean and covariance respectively, and use the mixed posterior mean and posterior covariance as the state information of the target.

[0019] A third aspect of this application provides a vehicle, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to perform the target tracking method as described in the above embodiments.

[0020] A fourth aspect of this application provides a computer-readable storage medium having a computer program or instructions stored thereon, which are executed by a processor to perform a target tracking method, such as the target tracking method.

[0021] Therefore, this application has at least the following beneficial effects:

[0022] This application's embodiments can calculate the prior probability and predicted value of a target under at least one motion model, and consider the target observation data at the current moment to calculate the target observation value under the motion model. Then, based on the observation value, predicted value, and prior probability, the posterior probability of the target under at least one motion model is calculated. Based on the posterior probability, the target's state information is calculated. Furthermore, it is not limited to a single motion model, allowing for more accurate determination of the target's state information, thereby improving the comfort and safety of subsequent vehicle operation. This solves the technical problem in related technologies where the accuracy of target state information prediction is low, thus affecting driving safety and comfort.

[0023] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0024] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0025] Figure 1 This is a flowchart of a target tracking method provided according to an embodiment of this application;

[0026] Figure 2 This is a flowchart illustrating the specific execution of the target tracking method provided in the embodiments of this application;

[0027] Figure 3 This is a flowchart of multi-motion model interactive tracking provided according to an embodiment of this application;

[0028] Figure 4 This is a flowchart of target filter management provided according to an embodiment of this application;

[0029] Figure 5 This is an example diagram of a target tracking device provided according to an embodiment of this application;

[0030] Figure 6 This is a structural schematic diagram of a vehicle provided according to an embodiment of this application. Detailed Implementation

[0031] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0032] Before describing the solution in this application, let's first introduce the current mainstream driving perception fusion method based on Kalman filtering, whose main inputs are:

[0033] 1. Vehicle motion status: Motion status information such as vehicle speed, acceleration, and angular velocity calculated based on wheel speed pulses, IMU (Inertial Measurement Unit), etc.

[0034] 2. Visual perception: At least one single forward-looking wide-angle camera, configurable forward-looking narrow-angle camera, panoramic camera, and rear-view camera, to obtain visual dynamic targets and visual lane lines through visual perception and post-processing.

[0035] 3. Millimeter-wave radar: Configurable single front radar or multiple corner radars to obtain target-level millimeter-wave radar observations.

[0036] 4. LiDAR: Configurable single or multiple lidar units to achieve target-level millimeter-wave radar observation.

[0037] Its main steps are:

[0038] 1. Calibration of different sensors, and synchronization of observation time and space.

[0039] 2. For targets that have been tracked, the mean and covariance are predicted based on a preset constant acceleration motion model.

[0040] 3. Based on algorithms such as greedy algorithms and Hungarian matching, the tracking target and the observations are matched, and the optimal observations are selected from multi-source observations.

[0041] 4. Track the updates of the mean and covariance of the target.

[0042] 5. Clearing out tracked targets that have lost observation and establishing tracking for newly observed targets.

[0043] Its output is:

[0044] A validated dynamic target with smooth motion and stable category attributes.

[0045] Furthermore, due to the use of a fixed acceleration motion model, when facing motor vehicles, such as braking vehicles, their actual acceleration changes rapidly, while the acceleration of the track (tracking target) target needs time to converge, resulting in lag. In scenarios such as ramps or curves with large curvature, vehicle motion is closer to nonlinear circular motion, and the tracking method using a linear model with a fixed acceleration motion model will also have errors that cannot be eliminated.

[0046] To address this issue, this application provides a target tracking method. In this method, the prior probability and predicted value of a target under at least one motion model are calculated. The target observation data at the current moment are considered in calculating the target observation value under the motion model. Then, based on the observation value, predicted value, and prior probability, the posterior probability of the target under at least one motion model is calculated. Based on the posterior probability, the target's state information is calculated, and the method is not limited to a single motion model. This solves the problem in related technologies where the accuracy of target state information prediction is low, thus affecting driving safety and comfort.

[0047] Specifically, Figure 1 This is a flowchart illustrating a target tracking method provided in an embodiment of this application.

[0048] like Figure 1 As shown, the target tracking method includes the following steps:

[0049] In step S101, the targets that have been tracked around the vehicle are acquired.

[0050] The target can be vehicles, objects, etc., around the vehicle.

[0051] In step S102, the prior probability of the target under at least one motion model is calculated, and the predicted value of the target under at least one motion model is calculated.

[0052] Among them, the motion models include CA (Constant Acceleration), CT (Constant Tangential Acceleration), and CV (Constant Velocity).

[0053] It is understood that the embodiments of this application can calculate the prior probability of the target under at least one motion model and calculate the predicted value of the target under at least one motion model so as to calculate the posterior probability subsequently.

[0054] It should be noted that the prior probability can be used to represent the likelihood of a certain model or state before considering the current observation data. For example, in the embodiments of this application, it is used to represent the probability that the target is in a certain motion model without considering the observation data at the current moment. The posterior probability is the result of updating the probability distribution of a certain event or parameter after obtaining some observation data. It is a probability calculated based on the prior probability and the newly obtained data (observation data).

[0055] The formula for calculating the prior probability in this application embodiment can be:

[0056]

[0057] in, Let be the prior probability value of the j-th motion model of the track target. Let M be the posterior probability of the i-th motion model of the track target at the previous time step. ij Let be the probability transition value from motion model i to motion model j in the Markov matrix.

[0058] In this application, different methods can be used for prediction based on different motion models. If it is linear, such as a uniform acceleration motion model, the general form of the KF (Kalman Filter) prediction formula can be used directly. If it is nonlinear, such as a uniform circular motion model, the Jacobian of the state transition needs to be calculated first, and then linearized before calculation, i.e., the EKF (Extended Kalman Filter) can be used. Alternatively, the probability distribution can be calculated after the sampling points are predicted, i.e., the UKF (Unscented Kalman Filter) can be used.

[0059] In this embodiment of the application, before calculating the prior probability of the target under at least one motion model, the method further includes: obtaining the current vehicle information, lane information, and the relative relationship between the vehicle and the target; and updating the target's motion model based on at least one of the current vehicle information, lane information, and the relative relationship between the vehicle and the target.

[0060] It is understood that the embodiments of this application can obtain at least one motion model of the target by acquiring the vehicle information, lane information, and the relative relationship between the vehicle and the target at the current moment.

[0061] In this embodiment, updating the target's motion model based on at least one of the following: current vehicle information, lane information, and the relative relationship between the vehicle and the target, includes: if the lane information indicates no lane lines, the target's motion model is a uniform acceleration motion model; if the lane information indicates the lane curvature is greater than a preset value, a uniform circular motion model is added to the target's motion model; if the vehicle information indicates the vehicle is waiting at an intersection and the target is on the vehicle's lane-changing side, a uniform circular motion model is added to the target's motion model; if the distance between the target and the vehicle meets a preset condition, a uniform circular motion model or a uniform linear motion model is added to the target's motion model.

[0062] The preset conditions can be set according to specific circumstances, such as the new target being the closest or the second closest to the vehicle.

[0063] It is understood that the embodiments of this application can update the motion model of the target, which can also be referred to as management, specifically as follows:

[0064] 1. Determine the presence of lane lines. If there are no lane lines, remove all filters from all track targets except for the CA model.

[0065] 2. Determine lane curvature. In high curvature scenes, add a CT model filter to all track targets.

[0066] 3. Determine the lane-changing status of the vehicle. If the vehicle is detected to be changing lanes, a CT model filter is added to the lane-changing track.

[0067] 4. Determine if the vehicle is waiting at an intersection. After detecting a waiting state at an intersection, add CT model filters to targets passing on both sides of the vehicle.

[0068] 5. Traverse all tracks to select the target. Add CT / CV model filters for CIPV (Closest In Path Vehicle) and Second CIPV (Second Closest In Path Vehicle).

[0069] In step S103, the predicted value and the target observation data at the current time are correlated to obtain the target observation value under at least one motion model, and the posterior probability of the target under at least one motion model is calculated based on the observation value, the predicted value and the prior probability.

[0070] The target observation data includes at least one of the following: visual dynamic target information, millimeter-wave radar target information, vehicle odometer information, and lidar target information.

[0071] It is understood that the embodiments of this application can associate the predicted value and the target observation data at the current time to obtain the observation value of the target under at least one motion model, and calculate the posterior probability of the target under at least one motion model based on the observation value, the predicted value and the prior probability. The specific calculation process is as follows.

[0072] In this embodiment of the application, obtaining the target's observation value under at least one motion model by associating the predicted value with the target observation data at the current time includes: synthesizing the predicted value into a mixed predicted value based on the prior probability; and associating the mixed predicted value with the target observation data at the current time to obtain the target's observation value under at least one motion model.

[0073] It is understood that, according to the embodiments of this application, predicted values ​​can be synthesized into mixed predicted values ​​based on prior probabilities, and the mixed predicted values ​​can be correlated with the target observation data at the current time to obtain the target observation value under at least one motion model. The mixing method of the mixed predicted values ​​is as follows:

[0074]

[0075] in, This is a mixed prediction mean vector. Let be the prior probability value of the i-th motion model. Let be the predicted mean vector of the i-th motion model.

[0076] Furthermore, algorithms such as greedy algorithms, Hungarian matching, and joint probability density association can be used to correlate predicted values ​​with target observation data at the current moment.

[0077] In this embodiment of the application, the posterior probability of the target under at least one motion model is calculated based on the observed value, the predicted value, and the prior probability, including: calculating the likelihood value between the observed value and the predicted value; and calculating the posterior probability of the target under at least one motion model based on the likelihood value and the prior probability.

[0078] It is understood that embodiments of this application can calculate the likelihood value between the observed value and the predicted value, and calculate the posterior probability of the target under at least one motion model based on the likelihood value and the prior probability, wherein,

[0079] The process of calculating the likelihood value is as follows:

[0080]

[0081] Among them, y i Let z be the residual vector of the i-th motion model, z be the observation vector, and H be the residual vector of the i-th motion model. i Let be the observation matrix of the i-th motion model. Let be the predicted mean vector of the i-th motion model.

[0082]

[0083] Where S i Let be the observation variance matrix of the i-th motion model. Let R be the prediction covariance matrix of the i-th motion model. i Let be the observation noise matrix of the i-th motion model.

[0084]

[0085] Among them, L i Let N be the likelihood value between the i-th motion model and the observed value. i Let be the dimension of the i-th motion model.

[0086] The formula for calculating the posterior probability is:

[0087]

[0088] in, Let be the posterior probability value of the j-th motion model. Let be the prior probability value of the i-th motion model. Let be the likelihood value of the i-th motion model.

[0089] In step S104, the target's state information is calculated based on the posterior probability to complete the tracking of the target.

[0090] It is understood that the embodiments of this application can calculate the target's state information based on the posterior probability to complete the tracking of the target. The target observation data at the current moment can be added to calculate a more accurate target state information.

[0091] In this embodiment of the application, calculating the target's state information based on the posterior probability includes: calculating the mean and covariance of the target's posterior probability under all motion models; calculating the mixed posterior mean and posterior covariance based on the mean and covariance respectively, and using the mixed posterior mean and posterior covariance as the target's state information.

[0092] It is understood that the embodiments of this application can calculate the mean and covariance of the posterior probability of the target under all motion models, and calculate the mixed posterior mean and posterior covariance based on the mean and covariance respectively, and use the mixed posterior mean and posterior covariance as the state information of the target, specifically referred to as:

[0093] By weighting the mean and covariance of each motion model of the target, the mixed posterior mean and covariance are obtained.

[0094]

[0095] in, This is the mixture of posterior mean vectors. Let be the posterior mean vector of the i-th motion model.

[0096]

[0097] in, The mixed posterior covariance matrix, Let be the posterior covariance matrix of the i-th motion model.

[0098] The mixed posterior mean and covariance are output as the current state of the tracked target and assigned to each motion model of the target.

[0099] It should be noted that the embodiments of this application can traverse each established tracking target and manage the target, such as clearing targets that have lost observation and establishing tracking for newly observed targets. By default, newly added targets only have the CA model set.

[0100] The target tracking method of this application is illustrated below through a specific embodiment, which mainly includes two modules: a multi-motion model interaction (IMM) tracking module and a track target filter management module.

[0101] The multi-motion model interactive tracking module is used to receive visual dynamic target information, millimeter-wave radar target information, vehicle odometer information, and lidar target information (optional), and output the tracked dynamic target. The internal track target can have multiple motion models at the same time, which are predicted and updated separately, and the output is the result after probability weighting. The filter configuration of each track target can change in real time.

[0102] The track target filter management module receives visual static target information, visual lane line information, vehicle odometer information, and fused track target information to determine the driving intentions of vehicles around the vehicle and manage the internal filter list of each track target within the multi-motion model interactive tracking algorithm.

[0103] The execution process of the method in this application is as follows: Figure 2 As shown:

[0104] 1. After the process is scheduled to be executed, newly received visual dynamic target information, millimeter-wave radar target information, vehicle odometer information, lidar target information (optional) and other messages are cached.

[0105] 2. Verify different sensors to determine if they are working properly and whether the delay has exceeded the timeout limit.

[0106] 3. To achieve time and space synchronization of observations from different sensors.

[0107] 4. Iterate through each track target that has been established and perform multi-motion model interactive tracking.

[0108] 5. During the multi-motion model interactive tracking process, the motion model of the track target is updated based on information such as lane line detection and relevant information of the vehicle, that is, the track target filter is managed, and finally the dynamic target is output.

[0109] Specifically:

[0110] 1. For each established track target, traverse the process and execute multi-motion model interactive tracking as follows: Figure 3 As shown, it includes the following steps:

[0111] 1. Based on the posterior probabilities of the previous frame or initialization on the track and the Markov transition matrix, calculate the prior probabilities of each motion model of the target on the current track. As follows:

[0112]

[0113] in, Let be the prior probability value of the j-th motion model of the track target. Let M be the posterior probability of the i-th motion model of the track target at the previous time step. ij Let be the probability transition value from motion model i to motion model j in the Markov matrix.

[0114] 2. Each motion model of the current track target makes predictions separately:

[0115] For different motion models, if it is linear, such as the CA constant acceleration motion model, the general form of the KF prediction formula can be used directly; if it is nonlinear, such as the CT uniform circular motion model, the Jacobian of the state transition needs to be calculated first, linearized, and then calculated, i.e., EKF is used, or the probability distribution is calculated after sampling point prediction, i.e., UKF.

[0116] Since the mixing operations of various models involve calculations between mean vectors, the state variables of various motion models in this algorithm are all 6-dimensional vectors, that is:

[0117] [xyvyaxay]T。 。

[0118] 3. The predicted values ​​of each motion model for the current track target are synthesized into a hybrid predicted value based on prior probabilities. This hybrid predicted value is then correlated with the sensor measurements of the current frame. The hybridization method is as follows:

[0119]

[0120] in, This is a mixed prediction mean vector. Let be the prior probability value of the i-th motion model. Let be the predicted mean vector of the i-th motion model.

[0121] 4. Perform data association between the track and observations from different sensors. Algorithms such as greedy algorithms, Hungarian matching, and Joint Probability Density Association (JPDA) can be used for data association.

[0122] 5. Calculate the likelihood of the predicted values ​​of each motion model of the current track target with the observed values ​​obtained by association.

[0123]

[0124] Among them, y i Let z be the residual vector of the i-th motion model, z be the observation vector, and H be the residual vector of the i-th motion model. i Let be the observation matrix of the i-th motion model. Let be the predicted mean vector of the i-th motion model.

[0125]

[0126] Among them, S i Let be the observation variance matrix of the i-th motion model. Let R be the prediction covariance matrix of the i-th motion model. i Let be the observation noise matrix of the i-th motion model.

[0127]

[0128] Among them, L i Let N be the likelihood value between the i-th motion model and the observed value. i Let be the dimension of the i-th motion model.

[0129] 6. The motion models of the current track target are updated using the correlation-obtained observations.

[0130] 7. Using the prior and likelihood of each motion model of the current track target obtained in steps 1 and 4, calculate the posterior probability of each motion model of the current track target and normalize it.

[0131]

[0132] in, Let be the posterior probability value of the j-th motion model. Let be the prior probability value of the i-th motion model. Let be the likelihood value of the i-th motion model.

[0133] 8. Using the posterior probabilities of each motion model of the current track target, weight the mean and covariance of each motion model of the current track target to obtain the mixed posterior mean and covariance.

[0134]

[0135] in, This is the mixture of posterior mean vectors. Let be the posterior mean vector of the i-th motion model.

[0136]

[0137] in, The mixed posterior covariance matrix, Let be the posterior covariance matrix of the i-th motion model.

[0138] 9. The mixed posterior mean and covariance are output as the state of the current track target and assigned to each motion model of the current track target.

[0139] 10. Traverse each established track target and manage the track targets. Clear track targets that have lost observation, create new tracks for newly observed targets, and by default, only set the CA model for new track targets.

[0140] Second, iterate through each established track target and manage the internal filters of the track targets, such as... Figure 4 As shown.

[0141] 1. Determine the presence of lane lines. If there are no lane lines, remove all filters from all track targets except for the CA model.

[0142] 2. Determine lane curvature. In high curvature scenes, add a CT model filter to all track targets.

[0143] 3. Determine the lane-changing status of the vehicle. If the vehicle is detected to be changing lanes, a CT model filter is added to the lane-changing track.

[0144] 4. Determine if the vehicle is waiting at an intersection. After detecting a waiting state at an intersection, add CT model filters to targets passing on both sides of the vehicle.

[0145] 5. Iterate through all tracks to select the target. For CIPV and Second CIPV, add a CT / CV model filter.

[0146] This application implements a multi-sensor fusion dynamic target tracking method based on the interaction of various vehicle motion models, including uniform acceleration model, uniform circular motion model, and uniform linear motion model. It can identify and track multiple types of dynamic and static targets, and filter their motion attributes such as position, speed, and acceleration. It can effectively reduce the speed and acceleration errors of moving targets in scenarios such as intersections, curves, and multi-vehicle road sections, provide more accurate target motion attributes, and improve the comfort of driving assistance functions in following other vehicles, starting, curves, and lane changes.

[0147] Compared to tracking algorithms that use a single motion model, the method in this application can more promptly converge the tracked state variables to near the true value when the surrounding target, such as a vehicle, is maneuvering (e.g., from going straight to turning, from moving to stationary, etc.); it also supports real-time configuration of the motion model, avoiding excessive computational consumption of the algorithm; and the motion model of the tracked target is switched adaptively, avoiding the addition of too much rule-based manual logic.

[0148] According to the target tracking method proposed in the embodiments of this application, the prior probability and predicted value of the target under at least one motion model can be calculated, and the target observation data at the current moment can be considered to calculate the target observation value under the motion model. Then, based on the observation value, predicted value and prior probability, the posterior probability of the target under at least one motion model can be calculated, and the state information of the target can be calculated based on the posterior probability. Moreover, it is not limited to a single motion model, and the state information of the target can be determined more accurately, thereby improving the comfort and safety of the vehicle in the subsequent driving process.

[0149] Next, the target tracking device proposed according to the embodiments of this application is described with reference to the accompanying drawings.

[0150] Figure 5 This is a block diagram of a target tracking device according to an embodiment of this application.

[0151] like Figure 5 As shown, the target tracking device 10 includes: an acquisition module 100, a first calculation module 200, a second calculation module 300, and a third calculation module 400.

[0152] The acquisition module 100 is used to acquire targets that have been tracked around the vehicle; the first calculation module 200 is used to calculate the prior probability of the target under at least one motion model and calculate the predicted value of the target under at least one motion model; the second calculation module 300 is used to correlate the predicted value and the target observation data at the current time to obtain the observation value of the target under at least one motion model, and calculate the posterior probability of the target under at least one motion model based on the observation value, the predicted value and the prior probability; the third calculation module 400 is used to calculate the state information of the target based on the posterior probability to complete the tracking of the target.

[0153] In this embodiment, the motion model includes a uniform acceleration motion model, a uniform circular motion model, and a uniform linear motion model, and the target observation data includes at least one of visual dynamic target information, millimeter-wave radar target information, vehicle odometer information, and lidar target information.

[0154] In this embodiment of the application, the apparatus 10 further includes an update module.

[0155] The update module is used to obtain the vehicle information, lane information, and relative relationship between the vehicle and the target at the current moment before calculating the prior probability of the target under at least one motion model; and to update the target's motion model based on at least one of the vehicle information, lane information, and relative relationship between the vehicle and the target at the current moment.

[0156] In this embodiment, the update module is further configured to: if the lane information indicates that there are no lane lines, then the motion model of the target is a uniform acceleration motion model; if the lane information indicates that the lane curvature is greater than a preset value, then a uniform circular motion model is added to the motion model of the target; if the vehicle information indicates that the vehicle is in a waiting state at an intersection and the target is on the lane-changing side of the vehicle, then a uniform circular motion model is added to the motion model of the target; if the distance between the target and the vehicle meets a preset condition, then a uniform circular motion model or a uniform linear motion model is added to the motion model of the target.

[0157] In this embodiment of the application, the second calculation module 300 is further configured to: synthesize a mixed prediction value based on the prior probability; and correlate the mixed prediction value with the target observation data at the current time to obtain the target observation value under at least one motion model.

[0158] In this embodiment of the application, the second calculation module 300 is further configured to: calculate the likelihood value between the observed value and the predicted value; and calculate the posterior probability of the target under at least one motion model based on the likelihood value and the prior probability.

[0159] In this embodiment of the application, the third calculation module 400 is further configured to: calculate the mean and covariance of the posterior probability of the target under all motion models; calculate the mixed posterior mean and posterior covariance based on the mean and covariance respectively, and use the mixed posterior mean and posterior covariance as the state information of the target.

[0160] It should be noted that the foregoing explanation of the target tracking method embodiment also applies to the target tracking device of this embodiment, and will not be repeated here.

[0161] According to the target tracking device proposed in the embodiments of this application, it can calculate the prior probability and predicted value of the target under at least one motion model, and take the target observation data at the current moment into account to calculate the target observation value under the motion model. Then, based on the observation value, predicted value and prior probability, it calculates the posterior probability of the target under at least one motion model, and calculates the target state information based on the posterior probability. Moreover, it is not limited to a single motion model, and can more accurately determine the target state information, thereby improving the comfort and safety of the vehicle during subsequent driving.

[0162] This application also provides a computer-readable storage medium storing a computer program or instructions thereon, which, when executed by a processor, implements the target tracking method described above.

[0163] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0164] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0165] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0166] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or more of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0167] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it includes one or a combination of the steps of the method embodiments.

Claims

1. A target tracking method characterized by, The method comprises the following steps: acquiring a target around the vehicle that has established tracking; calculating a prior probability of the target under at least one motion model and calculating a predicted value of the target under the at least one motion model, wherein the motion model comprises a uniform acceleration motion model, a uniform circular motion model and a uniform linear motion model; before calculating the prior probability of the target under the at least one motion model, the method further comprises: acquiring vehicle information, lane information and a relative relationship between the vehicle and the target at a current time; and updating the motion model of the target based on at least one of the vehicle information, the lane information and the relative relationship between the vehicle and the target at the current time; the updating of the motion model of the target based on at least one of the vehicle information, the lane information and the relative relationship between the vehicle and the target at the current time comprises: if the lane information is that there is no lane line, the motion model of the target is the uniform acceleration motion model; if the lane information is that the lane curvature is greater than a preset value, the uniform circular motion model is added to the motion model of the target; if the vehicle information is that the vehicle is in a waiting state at an intersection and the target is on a lane-changing side of the vehicle, the uniform circular motion model is added to the motion model of the target; and if a distance between the target and the vehicle satisfies a preset condition, the uniform circular motion model or the uniform linear motion model is added to the motion model of the target; associating the predicted value with target observation data at the current time to obtain an observation value of the target under the at least one motion model, and calculating a posterior probability of the target under the at least one motion model based on the observation value, the predicted value and the prior probability, wherein the target observation data comprises at least one of visual dynamic target information, millimeter wave radar target information, vehicle odometer information and laser radar target information; calculating state information of the target based on the posterior probability to complete tracking of the target.

2. The object tracking method of claim 1, wherein, The association of the predicted value with the target observation data at the current time to obtain the observation value of the target under the at least one motion model comprises: combining the predicted value with the prior probability to obtain a mixed predicted value; associating the mixed predicted value with the target observation data at the current time to obtain the observation value of the target under the at least one motion model.

3. The target tracking method according to claim 2, characterized in that, The calculation of the posterior probability of the target under the at least one motion model based on the observation value, the predicted value and the prior probability comprises: calculating a likelihood value between the observation value and the predicted value; calculating the posterior probability of the target under the at least one motion model based on the likelihood value and the prior probability.

4. The target tracking method according to claim 3, characterized by, The calculation of the state information of the target based on the posterior probability comprises: calculating a mean value and a covariance of the posterior probabilities of the target under all motion models; calculating a mixed posterior mean value and a posterior covariance based on the mean value and the covariance respectively, and taking the mixed posterior mean value and the posterior covariance as the state information of the target.

5. A target tracking device, characterized by, The method comprises the following steps: an acquiring module, configured to acquire a target around the vehicle that has established tracking; The first calculation module is configured to calculate a prior probability of the target under at least one motion model and to calculate a predicted value of the target under the at least one motion model, wherein the motion model comprises a uniform acceleration motion model, a uniform circular motion model, and a uniform linear motion model; before calculating the prior probability of the target under the at least one motion model, the method further comprises: obtaining ego vehicle information, lane information, and a relative relationship between the ego vehicle and the target at a current time; and updating the motion model of the target based on at least one of the ego vehicle information, the lane information, and the relative relationship between the ego vehicle and the target; the updating of the motion model of the target based on at least one of the ego vehicle information, the lane information, and the relative relationship between the ego vehicle and the target comprises: if the lane information indicates that there is no lane line, the motion model of the target is the uniform acceleration motion model; if the lane information indicates that a lane curvature is greater than a preset value, the uniform circular motion model is added to the motion model of the target; if the ego vehicle information indicates that the ego vehicle is in a waiting state at an intersection and the target is on a lane-changing side of the ego vehicle, the uniform circular motion model is added to the motion model of the target; and if a distance between the target and the ego vehicle satisfies a preset condition, the uniform circular motion model or the uniform linear motion model is added to the motion model of the target; The second calculation module is configured to associate the predicted value with target observation data at the current time to obtain an observation value of the target under the at least one motion model, and to calculate a posterior probability of the target under the at least one motion model based on the observation value, the predicted value, and the prior probability, wherein the target observation data comprises at least one of visual dynamic target information, millimeter wave radar target information, ego vehicle odometer information, and laser radar target information. The third calculation module is configured to calculate state information of the target based on the posterior probability, so as to complete tracking of the target.

6. A vehicle characterized by comprising: The computer program or instructions are executed by the processor to implement the target tracking method according to any one of claims 1-4. The computer program or instructions are executed by the processor to implement the target tracking method according to any one of claims 1-4.

7. A computer readable storage medium having stored thereon a computer program or instructions, characterized in that, ​

Images