Course angle determination method, controller, vehicle, and program product

Abstract

The present disclosure relates to a heading angle determination method, a controller, a vehicle and a program product, and relates to the technical field of automatic driving. The heading angle determination method comprises: obtaining motion information of a visual target and an observed value of a heading angle of the visual target in a current frame in a pure visual perception scene; determining a predicted value of the heading angle according to the motion information; determining a confidence level of the observed value of the heading angle according to a spatial position of the visual target in a coordinate system of a host vehicle; and determining a target heading angle according to the confidence level, the observed value of the heading angle and the predicted value of the heading angle.

Description

Technical Field

[0001] This disclosure relates to the field of autonomous driving technology, specifically to a heading angle determination method, a controller, a vehicle, and a software product. Background Technology

[0002] In related technologies, the determination of heading angle in pure vision autonomous driving systems is highly dependent on image appearance features and the generalization ability of the learning model. However, the image appearance features of visual targets are affected by changes in illumination, occlusion, viewpoint, and target symmetry. When the movement speed of the visual target is low or the attitude change is not obvious, the heading angle output by the learning model based on the image appearance features often has problems such as instability, jitter, or directional ambiguity. Summary of the Invention

[0003] To address the shortcomings of related technologies, this disclosure provides a heading angle determination method, controller, vehicle, and program product.

[0004] To achieve the above objectives, in a first aspect, this disclosure provides a method for determining a heading angle, comprising: Acquire motion information of the visual target and the heading angle observation value of the visual target in the current frame under a pure visual perception scene; Based on the motion information, determine the predicted heading angle; The confidence level of the heading angle observation value is determined based on the spatial position of the visual target in the vehicle coordinate system; The target heading angle is determined based on the confidence level, the observed heading angle, and the predicted heading angle.

[0005] Secondly, this disclosure provides a controller, including: The first memory, on which the computer program is stored; A first processor is configured to execute the computer program in the first memory to implement the method described in the first aspect.

[0006] Thirdly, this disclosure provides a vehicle including the controller described in the second aspect.

[0007] Fourthly, this disclosure provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method described in the first aspect.

[0008] Fifthly, this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0009] By constraining the heading angle prediction value with the motion information of the visual target, the heading angle prediction value can be dynamically adjusted according to the actual motion state of the visual target, avoiding the prediction deviation caused by the fixed process noise assumption in traditional methods, and improving the accuracy and reliability of the heading angle prediction value. Based on the spatial position of the visual target in the vehicle coordinate system, the confidence level of the heading angle observation value is determined, and differentiated observation confidence strategies can be adopted for visual targets at different distances, thereby improving the overall estimation accuracy. Based on the confidence level, the heading angle observation value and the heading angle prediction value, the target heading angle is determined. The entire process effectively constrains the heading angle estimation process from two levels: physical motion law and spatial geometric relationship, significantly improving the stability and accuracy of heading angle estimation in pure visual perception scenarios.

[0010] Other features and advantages of this disclosure will be described in detail in the following detailed description section. Attached Figure Description

[0011] The accompanying drawings are provided to further illustrate the present disclosure and form part of the specification. They are used together with the following detailed description to explain the present disclosure, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating a heading angle determination method according to an exemplary embodiment of the present disclosure.

[0012] Figure 2 This is another flowchart illustrating a heading angle determination method according to an exemplary embodiment of the present disclosure.

[0013] Figure 3 This is a block diagram of a controller according to an exemplary embodiment of the present disclosure.

[0014] Figure 4 This is a block diagram of a vehicle according to an exemplary embodiment of the present disclosure. Detailed Implementation

[0015] The specific embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit this disclosure.

[0016] It is worth noting that in autonomous driving systems, the environmental perception module needs to accurately perceive the spatial position, motion state, and attitude information of surrounding traffic participants. Among these, the target's heading angle describes the target's orientation on the ground plane and is one of the important fundamental parameters for target tracking, trajectory prediction, and decision-making. Especially in pure vision perception solutions that rely solely on onboard cameras, how to stably and accurately obtain the target's heading angle has always been a key technical challenge in the field of perception.

[0017] As mentioned in the background section, in related technologies, pure vision-based autonomous driving systems typically process image data acquired by cameras using deep learning models. They detect targets through neural networks and simultaneously output the target's spatial position, size, and heading angle information in the detection results. The target heading angle is generally estimated using regression or by parameterizing the angle to avoid discontinuities. However, in practical engineering applications, to improve the temporal continuity of multi-frame output results, state estimation algorithms such as Kalman filtering are often introduced based on the detection results to smooth and predict the target heading angle.

[0018] However, due to the lack of direct 3D geometric constraints in purely visual solutions, the estimation of the target heading angle is highly dependent on image appearance features and the generalization ability of the learning model, making it susceptible to factors such as illumination changes, occlusion, viewpoint changes, and target symmetry. When the target's movement speed is low or its attitude change is not significant, the heading angle results output by deep learning models based on image appearance features often suffer from instability, jitter, or directional ambiguity. Furthermore, traditional Kalman filtering, when estimating the heading angle, typically assumes that the target's orientation changes follow a linear or approximately linear motion model. This assumption is difficult to hold in complex traffic scenarios over long periods, resulting in the filtered heading angle results still exhibiting jumps or accumulated errors.

[0019] In view of this, this disclosure provides a heading angle determination method, controller, vehicle, and program product, which effectively constrains the heading angle estimation process based on physical motion laws and spatial geometric relationships, thereby improving the accuracy and reliability of the heading angle.

[0020] It is worth noting that the heading angle determination method provided in this disclosure takes the vehicle autonomous driving scenario as the main example, but it is also applicable to other purely visual perception scenarios, such as UAV visual tracking and target attitude estimation in intelligent monitoring systems. As long as the application involves estimating the target heading angle from image sequences, it falls within the application scope of this disclosure.

[0021] Figure 1 This is a flowchart illustrating a heading angle determination method according to an exemplary embodiment of this disclosure, such as... Figure 1 As shown, the method for determining the heading angle may include the following steps: In step S11, the motion information of the visual target and the heading angle observation value of the visual target in the current frame under pure visual perception scene are obtained.

[0022] In step S12, the predicted heading angle is determined based on the motion information.

[0023] In step S13, the confidence level of the heading angle observation value is determined based on the spatial position of the visual target in the vehicle coordinate system.

[0024] In step S14, the target heading angle is determined based on the confidence level, the observed heading angle, and the predicted heading angle.

[0025] In this context, visual targets refer to traffic participants detected in images captured by the vehicle-mounted camera, including but not limited to vehicles traveling ahead and pedestrians on the roadside. Motion information refers to the motion state data of the targets accumulated in historical frames, including but not limited to the target's velocity vector and acceleration in the ground plane. Each visual target in each frame has a corresponding heading angle observation value.

[0026] The vehicle coordinate system is a coordinate system established with the vehicle as the origin. Typically, the forward direction of the vehicle is defined as the positive x-axis, and the left side as the positive y-axis. The spatial position of a visual target within the vehicle coordinate system can be determined through 3D detection or calculated using methods such as monocular depth estimation.

[0027] It is worth noting that the process of determining the predicted heading angle involves prediction uncertainty, which characterizes the possible range of variation of the predicted heading angle between adjacent frames. If the prediction uncertainty is set too high, the constraint on the heading angle variation is too loose, easily leading to jitter in the estimation results due to noise interference; if the prediction uncertainty is set too low, the constraint on the heading angle variation is too tight, making it impossible to track the visual target in a timely manner when a large heading angle change actually occurs. Therefore, the prediction uncertainty can be dynamically constrained through motion information, enabling the prediction process to adaptively match the actual motion characteristics of the visual target, allowing the predicted heading angle value to be dynamically adjusted according to the actual motion state of the visual target. For example, when the visual target is moving at a high speed, its possible heading angle change per unit time is large, and the corresponding prediction uncertainty is also large; when the target is stationary or at low speed, the range of heading angle change is small, and the prediction uncertainty is correspondingly reduced.

[0028] It should be understood that determining the heading angle observation involves observation uncertainty, which characterizes the confidence level of the heading angle observation in the current frame. In a purely visual perception system, there is an inherent physical relationship between observation accuracy and the distance to the visual target: visual targets closer to the vehicle occupy a larger pixel area in the image, have richer appearance features and stronger geometric constraints, resulting in higher reliability of the heading angle observation, i.e., higher confidence level. Conversely, targets farther away occupy only a few pixels in the image, have blurred appearance information, and significantly increase the uncertainty of the heading angle observation, i.e., lower confidence level. Therefore, determining the confidence level of the heading angle observation based on its spatial location allows for differentiated observation confidence strategies for targets at different distances.

[0029] In the above technical solution, in the prediction dimension, constraining the heading angle prediction value by the motion information of the visual target allows the heading angle prediction value to be dynamically adjusted according to the actual motion state of the visual target. This makes the prediction process conform to the real motion characteristics of the target, avoiding the prediction deviation caused by fixed process noise assumptions in traditional methods, and improving the accuracy and reliability of the heading angle prediction value. In the observation dimension, determining the confidence level of the heading angle observation value based on the spatial position of the visual target in the vehicle coordinate system allows for differentiated observation confidence strategies for visual targets at different distances, making the confidence level match the inherent accuracy characteristics of pure visual perception, thereby improving the overall estimation accuracy. Based on the confidence level, the heading angle observation value, and the heading angle prediction value, the target heading angle is determined. The entire process effectively constrains the heading angle estimation process from two levels: physical motion laws and spatial geometric relationships. This ensures that the heading angle estimation process maintains a reasonable constraint strength under different motion states and observation conditions, effectively improving the stability and accuracy of heading angle estimation in pure visual perception scenarios.

[0030] To facilitate a better understanding of the heading angle determination method provided in this disclosure by those skilled in the art, the heading angle determination method will be described in detail below.

[0031] In one feasible embodiment, the motion information includes velocity and the time interval between adjacent frames; in step S12, determining the predicted heading angle based on the motion information may include: The range of heading angle variation corresponding to the visual target is determined based on the speed and the time interval between adjacent frames; The range of heading angle variation is used as a process disturbance term, and the predicted heading angle value is determined based on the process disturbance term and the historical heading angle.

[0032] It's worth noting that velocity refers to the magnitude of the velocity vector of the visual target in the ground plane. This can be calculated by differential calculation based on the positional information accumulated in historical frames, or it can be obtained directly from the velocity estimate output by the target detection model. The adjacent frame time interval refers to the time difference between two adjacent image frames. In scenarios where vehicle-mounted cameras acquire images at a fixed frame rate, this time interval is usually a fixed value; for example, at a acquisition frequency of 30 frames per second, the adjacent frame time interval is approximately 33 milliseconds. The historical heading angle refers to the target heading angle in historical frames within a purely visual perception scenario. The predicted heading angle value refers to the predicted heading angle of the visual target in the current frame within a purely visual perception scenario.

[0033] In this embodiment, the velocity v of the visual target can be determined based on the motion information of the visual target in historical frames. k =‖v xy k Let Δt be the time interval between adjacent frames. Substituting the velocity and the time interval between adjacent frames into the following formula, we can obtain the range of change in the heading angle of the visual target in the current frame. θ max : , Here, f() represents the mapping function set according to the target's motion characteristics.

[0034] range of heading angle variation θ max w as a process disturbance term k The process disturbance term w k and historical heading angle θ k-1 Substitute the values ​​into the following formula for time-based recursive prediction to determine the predicted heading angle θ. - k : .

[0035] In the above technical solution, the range of heading angle variation is used as a process disturbance term to constrain the heading angle prediction value. The process disturbance term represents the range of heading angle variation during the heading angle prediction process.

[0036] In a feasible embodiment, in step S13, determining the confidence level of the heading angle observation value based on the spatial position of the visual target in the vehicle coordinate system may include: Determine the spatial region where the visual target is located based on its spatial position in the vehicle's coordinate system; The confidence level of the heading angle observation is determined based on the distance between the spatial area where the visual target is located and the vehicle. The confidence level is inversely proportional to the distance between the spatial area where the visual target is located and the vehicle.

[0037] It is worth noting that the spatial area where the visual target is located and the distance between the vehicle can be divided into multiple distance regions, with each distance region corresponding to a confidence level. The number of distance regions is not limited in this disclosure.

[0038] For example, when dividing the spatial region where the visual target is located and the distance between the vehicle into a near-distance region and a far-distance region, when the spatial region where the visual target is located is a near-distance region, the heading angle observation value is determined to correspond to a first confidence level; when the spatial region where the visual target is located is a far-distance region, the heading angle observation value is determined to correspond to a second confidence level; wherein, the first confidence level is greater than the second confidence level.

[0039] It's worth noting that the threshold for dividing the distance region can be preset based on sensor characteristics (such as camera resolution and focal length) and the performance of the target detection model. For example, the area within 50 meters in front of the vehicle can be defined as the near-distance region, and anything beyond 50 meters can be defined as the far-distance region. Alternatively, three or more spatial regions can be set, each corresponding to a different confidence level, to achieve more refined adjustment of the observation confidence. For example, a mid-distance region can be defined, corresponding to an intermediate confidence level between the first and second confidence levels.

[0040] In this embodiment, the spatial position of the visual target in the vehicle coordinate system is p. k =(x k ,y k ), where x k Let x be the coordinate of the visual target in the vehicle's coordinate system, and y be the coordinate of the target in the x-direction. k Let be the y-coordinate of the visual target in the vehicle coordinate system. If the visual target is determined to be in a preset near-range area based on its spatial location, then the heading angle observation value corresponds to a higher confidence level, indicating higher reliability, and a smaller observation noise is used. If the visual target is determined to be in a far-range area based on its spatial location, then the heading angle observation value corresponds to a lower confidence level, indicating lower reliability, and a larger observation noise is used. This constructs a heading angle observation noise model related to the spatial location of the visual target. , Among them, R k Characterizing the measurement noise covariance matrix, σ 2 near σ represents the variance of measurement noise when the visual target is in the static distance region. 2 far The variance of measurement noise characterizes the visual target when it is in a distant region.

[0041] In the above technical solution, when the visual target is in a close-range area, a first confidence level is selected, resulting in a larger gain for the heading angle state update. The heading angle observation value of the current frame is trusted, and the predicted value is significantly corrected, thus fully utilizing high-quality observation information to improve estimation accuracy. When the visual target is in a distant-range area, a second confidence level is selected, resulting in a smaller gain for the heading angle state update. The predicted value is maintained with only minor adjustments, avoiding excessive interference from low-quality observations on the estimation results. Adaptive observation noise modeling based on spatial location can dynamically adjust the weight of observation values ​​in the state update according to the inherent characteristic of observation accuracy decaying with distance in pure visual perception, significantly improving the heading angle estimation quality under different distance conditions.

[0042] In one feasible embodiment, the heading angle determination method may further include: Obtain historical heading angle observations of the visual target and determine the angle difference between the observed heading angles and the historical heading angle observations; If the angle difference meets the preset stability condition, the heading angle observation value is filtered and corrected according to the observation uncertainty to obtain the corrected heading angle observation value. If the angle difference does not meet the preset stability condition, the historical heading angle observation value of the most recent frame will be used as the heading angle observation value of the current frame.

[0043] It's worth noting that historical heading angle observations refer to the sequence of heading angle observations acquired in the preceding frames. Preset stability conditions are used to determine whether the heading angle observation in the current frame is consistent with the historical observation sequence. For example, an angle difference threshold can be set. When the angle difference between the current heading angle observation and the historical heading angle observations in the most recent frames is less than this threshold, the current observation is considered to meet the stability condition and is considered a normal observation. Conversely, if the angle difference exceeds the threshold, the current observation is considered to not meet the stability condition and is considered an abnormal observation. For example, if the heading angle of a vehicle ahead has remained relatively stable with slight fluctuations in the vicinity of directly in front for the past few frames, but suddenly jumps to pointing to the side in the current frame, this abrupt change is likely due to occlusion, sudden changes in lighting, or model false detection, rather than a genuine drastic change in the target's orientation.

[0044] In the above technical solution, normal and abnormal observations are distinguished by whether the angle difference between the observed heading angle and the historical heading angle meets the stability condition. For normal observations that meet the stability condition, the heading angle observed value is allowed to participate in the subsequent filtering and correction process to obtain the target heading angle. For abnormal observations that do not meet the stability condition, the historical heading angle observed value of the corresponding most recent frame is used as the target heading angle of the visual target.

[0045] In a feasible embodiment, if the angle difference does not meet a preset stability condition, the historical heading angle observation value corresponding to the most recent frame is used as the heading angle observation value of the current frame, which may include: If the angle difference is not within the preset range, an abnormal count of heading angle observations is performed. When the cumulative number exceeds the preset threshold, it is determined that the angle difference does not meet the preset stability condition. The historical heading angle observation of the corresponding most recent frame is used as the heading angle observation of the visual target, and the count is set to zero.

[0046] The preset range and preset threshold can be preset according to the noise level of the application scenario and the motion characteristics of the target, and this disclosure does not limit them.

[0047] It is worth noting that for abnormal observations where the angle difference is not within the preset range, the corresponding heading angle observation value is marked as abnormal and counted cumulatively. This effectively isolates the pollution of the heading angle estimation results by abnormal jumps in a single frame, avoiding situations where the estimation results deviate significantly from the true value due to a single erroneous observation. When the cumulative count of abnormal observations exceeds a preset threshold, it indicates that the current heading angle observation sequence has undergone a persistent unstable transition. This may be due to the target reappearing after being occluded for a long time, or a fundamental change in the target's appearance causing the model output to remain abnormal. At this point, instead of relying on the observation results of the current frame, the output is reverted to the stable historical heading angle observation value of the most recent frame, while the abnormal count is reset to zero, creating conditions for the re-stabilization of subsequent observation sequences.

[0048] In the above technical solution, the anomaly accumulation recovery mechanism achieves a reasonable balance between robustness and response speed. A single anomaly observation may only be an occasional noise interference; by ignoring this observation and continuing to wait for subsequent normal observations to recover, overreaction is avoided. However, when anomalies continue to accumulate, it indicates that the problem is no longer occasional; the mechanism reverts to a stable value and resets the state, preventing the estimation result from deviating from the true value for a long time. In this embodiment, the current frame heading angle observation value θ is... obs k Convert to unit direction vector h k =(cosθ obs k sinθ obs k ); The unit direction vector h k Add the observed heading angle to the historical heading angle observation sequence, while keeping the length of the historical sequence no more than a preset number of frames, and calculate the angle difference Δθ between the observed heading angle and the historical heading angle observation: Δθ=|wrap(θ obs k -θ obs k-i )|, Among them, wrap() represents the angle normalization function, which is used to restrict any angle value to a continuous interval of [-π, π).

[0049] Determine whether the angle difference meets the preset stability condition to characterize the continuity between multiple observation frames; if the angle difference meets the preset stability condition, then adjust θ according to the observation uncertainty. obs k Perform filtering correction to obtain the corrected heading angle observation value; if the angle difference is not within the preset range, mark the heading angle observation value θ. obs kFor abnormal observations, increment the count of abnormal observations by one. When the cumulative count exceeds a preset threshold, it is determined that the angle difference does not meet the preset stability condition. The historical heading angle observation value of the corresponding most recent frame is used as the heading angle observation value of the current frame, and the count is set to zero.

[0050] In a feasible embodiment, determining the target heading angle in step S14 based on the confidence level, the observed heading angle, and the predicted heading angle may include: Based on the confidence level, determine the state update gain between the observed and predicted heading angle values; Determine the residual between the observed and predicted heading angle values; The target heading angle is obtained based on the state update gain, the heading angle observation, and the residual.

[0051] It is worth noting that the state update gain characterizes the degree of confidence in the residuals, which are the differences between the observed and predicted heading angles. A larger state update gain results in a stronger correction to the residuals, and the final result is closer to the observed heading angle; conversely, a smaller state update gain results in a weaker correction to the residuals, and the final result is closer to the predicted heading angle.

[0052] It should be understood that the calculation of the state update gain depends on both the range of variation of the predicted heading angle and the confidence level of the observed heading angle. When the confidence level of the observed heading angle is low and the range of variation of the predicted heading angle is large, the gain is close to 1, indicating high confidence in the observed heading angle; when the confidence level of the observed heading angle is high and the range of variation of the predicted heading angle is small, the gain is close to 0, indicating high confidence in the predicted heading angle. In this embodiment, the confidence level of the observed heading angle determined based on spatial location in the previous embodiments is explicitly introduced into the gain calculation process, so that the gain can be adaptively adjusted as the target distance changes.

[0053] For example, the heading angle observation value θ is determined based on the observation uncertainty. obs k With the predicted heading angle θ pre k State update gain K k and residuals (θ) obs k -θ pre k ), and update the state gain K k Residual (θ) obs k -θ pre k and the observed heading angle θ obs k Substituting into the following formula, we obtain the target heading angle θ: θ=θpre k +K k (θ obs k -θ pre k ).

[0054] It's worth noting that the residuals between the predicted and observed heading angles are determined, and the confidence level is used as a weighting adjustment factor to determine the contribution ratio of the predicted and observed heading angles in the final estimation result. Specifically, when the confidence level is low, the system tends to trust the observed heading angles and make larger corrections to the predicted heading angles; when the confidence level is high, the system tends to maintain the predicted heading angles and make only minor adjustments. The final target heading angle is the optimal fusion result of the predicted and observed heading angles under adaptive weights, maintaining continuity in the time dimension while utilizing the observation information of the current frame for correction.

[0055] In one feasible embodiment, the heading angle determination method may further include: If the visual target is stationary or moving at low speed, and the number of historical heading angle observations of the visual target reaches a preset number, then the multiple historical heading angle observations are converted into direction vectors to obtain a vector set. Cluster analysis is performed on the vector set to obtain multiple directional clusters, and the centroid direction of the directional cluster with the largest number of samples is determined as the target heading angle.

[0056] The preset quantity can be preset according to the actual scene state or prediction accuracy requirements, and this disclosure does not limit it. The speed of the visual target can be used to determine whether the visual target is stationary or moving at low speed. This disclosure also does not limit it.

[0057] For example, if the visual target is stationary or moving at low speed, and the number of historical heading angle observations of the visual target reaches a preset number, then obtain the set of historical heading angle unit vectors H={h1,h2,…,h...}. n}, and perform cluster analysis on the set H, dividing it into k directional clusters, and then use the following formula to minimize the intra-cluster distance: , in, The centroid of the directional cluster to which the sample belongs.

[0058] Select the orientation cluster with the largest number of samples, and use the centroid orientation corresponding to the orientation cluster as the re-estimation result of the target heading angle: h cluster =μ max。

[0059] Based on the filtering update results or clustering reestimation results, the final heading angle of the target is determined: .

[0060] In the above technical solution, multiple historical observations are converted into direction vectors and smoothed in the time domain of the angle domain, eliminating false heading changes caused by low speed. Cluster analysis is performed on the vector geometry, and the centroid direction of the direction cluster with the most samples is determined as the target heading angle. The cluster analysis automatically identifies and removes isolated small clusters (i.e., outliers), retaining only the mainstream direction cluster with the most samples, which effectively prevents interference from short-term erroneous observations on the estimation results and improves the accuracy and reliability of the target heading angle.

[0061] The following describes the heading angle determination method provided in this disclosure using a complete embodiment, such as... Figure 2 As shown, the method for determining the heading angle may include the following steps: I. Determine the observed heading angle: i. Obtain the heading angle observation value of the visual target in the current frame under pure visual perception scene, as well as the historical heading angle observation value of the visual target.

[0062] ii. Determine the angle difference between the observed heading angle and the historical heading angle.

[0063] iii. Determine whether the angle difference meets the preset stability condition. If yes, proceed to step iv; otherwise, proceed to step v.

[0064] iv. Filter and correct the heading angle observations based on the confidence level to obtain the corrected heading angle observations.

[0065] v. Use the historical heading angle observation value corresponding to the most recent frame as the heading angle observation value of the current frame.

[0066] II. Determine the predicted heading angle: i. The speed at which visual targets are acquired and the time interval between adjacent frames.

[0067] ii. Determine the range of heading angle variation corresponding to the visual target based on the velocity and the time interval between adjacent frames.

[0068] iii. Treat the range of heading angle variation as a process disturbance term, and determine the predicted heading angle value based on the process disturbance term and the historical heading angle.

[0069] III. Calculate the target heading angle: i. Determine the spatial region where the visual target is located based on its spatial position in the vehicle coordinate system.

[0070] ii. Determine the relationship between the determined spatial region and the preset distance region. Proceed to steps iii-iv, or steps v-vi.

[0071] iii. If the spatial region is a near-distance region, determine the first confidence level corresponding to the heading angle state observation and execute step iv.

[0072] iv. Determine the target heading angle based on the first confidence level, the observed heading angle, and the predicted heading angle.

[0073] v. If the spatial region is a distant region, then determine the second confidence level corresponding to the heading angle state observation.

[0074] vi. Determine the target heading angle based on the second confidence level, the observed heading angle, and the predicted heading angle.

[0075] In the above technical solution, an adaptive modeling mechanism for the range of changes in the predicted heading angle driven by motion state and an adaptive modeling mechanism for the confidence level of the observed heading angle driven by spatial position are constructed. This jointly constrains the state prediction and observation update process of the heading angle from two dimensions: physical motion laws and spatial geometric relationships. On the one hand, the range of heading angle changes determined by the target velocity and the inter-frame time interval is used as a process disturbance term, ensuring that the predicted heading angle process closely matches the target's actual motion characteristics. Larger heading angle changes are allowed when the target is moving at high speed to ensure tracking response speed, while the prediction range is tightened when the target is moving at low speed or stationary to suppress noise propagation. This effectively overcomes the insufficient adaptability of traditional fixed process noise models in complex motion scenarios. On the other hand, different confidence levels are assigned to the near-distance and far-distance regions based on the target's spatial position. This adaptively adjusts the weight of the observed heading angle in the state update based on the inherent characteristic of observation accuracy decaying with distance in pure visual perception, avoiding excessive interference from low-quality observations at long distances on the estimation results. Furthermore, through consistency analysis of multi-frame historical observations and an anomaly observation accumulation counting mechanism, single-frame anomalous jumps can be effectively identified and isolated. Simultaneously, rapid recovery is achieved through backtracking to historical stable values ​​when anomalies accumulate continuously, further enhancing the robustness of heading angle estimation under adverse conditions such as occlusion and sudden changes in illumination. The clustering re-estimation mechanism for stationary or low-speed targets utilizes the physical prior of short-term constant heading angle from a statistical perspective, effectively resolving directional ambiguity and random drift problems when motion constraints are lacking.

[0076] Based on the same inventive concept, this disclosure also provides a controller, such as Figure 3 As shown, the controller includes: The first memory 301 stores computer programs; The first processor 302 is configured to execute the computer program in the first memory 301 to implement the above-described heading angle determination method.

[0077] The first processor 302 controls the overall operation of the controller 300 to complete all or part of the steps in the heading angle determination method described above. The first memory 301 stores various types of data to support the operation of the controller 300. This data may include, for example, instructions for any application or method operating on the controller 300, as well as application-related data, such as motion information of the visual target, the heading angle observation value of the visual target in the current frame in a purely visual perception scene, the heading angle prediction value, confidence level, etc. The first memory 301 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0078] In an exemplary embodiment, the controller 300 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the heading angle determination method described above.

[0079] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, which, when executed by a processor, implement the steps of the heading angle determination method described above. For example, the computer-readable storage medium may be the first memory 301 including the program instructions described above, which may be executed by the first processor 302 of the controller 300 to complete the heading angle determination method described above.

[0080] In another exemplary embodiment, a computer program product is also provided, which includes a computer program executable by a processor, which, when executed by the processor, implements the steps of the heading angle determination method described above.

[0081] Based on the same inventive concept, this disclosure also provides a vehicle including the aforementioned controller.

[0082] Figure 4 This is a block diagram illustrating a vehicle 400 according to an exemplary embodiment. For example, vehicle 400 can be a hybrid vehicle, a non-hybrid vehicle, an electric vehicle, a fuel cell vehicle, or other type of vehicle. Vehicle 400 can be an autonomous vehicle or a semi-autonomous vehicle. (Refer to...) Figure 4 The vehicle 400 may include various subsystems, such as an infotainment system 410, a perception system 420, a decision control system 430, a drive system 440, and a computing platform 450. The vehicle 400 may also include more or fewer subsystems, and each subsystem may include multiple components. Furthermore, each subsystem and component of the vehicle 400 can be interconnected via wired or wireless means. In some embodiments, the infotainment system 410 may include a communication system, an entertainment system, and a navigation system. The perception system 420 may include several sensors for sensing information about the environment surrounding the vehicle 400. For example, the perception system 420 may include a Global Positioning System (GPS, BeiDou, or other positioning systems), an Inertial Measurement Unit (IMU), lidar, millimeter-wave radar, ultrasonic radar, and a camera device. The decision control system 430 may include a computing system, a vehicle controller, a steering system, a throttle, a braking system, and the aforementioned controllers.

[0083] The drive system 440 may include components that provide powered motion to the vehicle 400. In one embodiment, the drive system 440 may include an engine, an energy source, a transmission system, and wheels. The engine may be one or a combination of internal combustion engines, electric motors, and compressed air engines. The engine is capable of converting energy provided by the energy source into mechanical energy. Some or all of the functions of the vehicle 400 are controlled by a computing platform 450. The computing platform 450 may include at least one second processor 451 and a second memory 452, the second processor 451 being capable of executing instructions 453 stored in the second memory 452. The second processor 451 may be any conventional processor, such as a commercially available CPU. The processor may also include, for example, a Graphics Processing Unit (GPU), a Field Programmable Gate Array (FPGA), a System on Chip (SOC), an Application Specific Integrated Circuit (ASIC), or a combination thereof. The second memory 452 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. In addition to instructions 453, the second memory 452 can also store data, such as road maps, route information, vehicle position, direction, speed, etc. The data stored in the second memory 452 can be used by the computing platform 450. In this embodiment of the disclosure, the second processor 451 can execute instructions 453 to complete all or part of the steps of the heading angle determination method described above, or complete all or part of the steps of the heading angle determination method described above through the controller in the decision control system 430.

[0084] The preferred embodiments of this disclosure have been described in detail above with reference to the accompanying drawings. However, this disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this disclosure, various simple modifications can be made to the technical solutions of this disclosure, and these simple modifications all fall within the protection scope of this disclosure.

[0085] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, this disclosure will not describe the various possible combinations separately.

[0086] Furthermore, various different embodiments of this disclosure can be combined in any way, as long as they do not violate the spirit of this disclosure, they should also be regarded as the content disclosed in this disclosure.

Claims

1. A method for determining a heading angle, characterized in that, include: Acquire motion information of the visual target and the heading angle observation value of the visual target in the current frame under a pure visual perception scene; Based on the motion information, determine the predicted heading angle; The confidence level of the heading angle observation value is determined based on the spatial position of the visual target in the vehicle coordinate system. The target heading angle is determined based on the confidence level, the observed heading angle, and the predicted heading angle.

2. The method for determining the heading angle according to claim 1, characterized in that, The motion information includes velocity and the time interval between adjacent frames; determining the predicted heading angle based on the motion information includes: The range of heading angle variation corresponding to the visual target is determined based on the speed and the time interval between adjacent frames; The range of heading angle variation is used as a process disturbance term, and the predicted heading angle value is determined based on the process disturbance term and the historical heading angle.

3. The method for determining the heading angle according to claim 1, characterized in that, The step of determining the confidence level of the heading angle observation value based on the spatial position of the visual target in the vehicle coordinate system includes: Based on the spatial position of the visual target in the vehicle coordinate system, determine the spatial region where the visual target is located; The confidence level of the heading angle observation is determined based on the distance between the spatial region where the visual target is located and the vehicle. The distance between the spatial region where the visual target is located and the vehicle is inversely proportional to the confidence level.

4. The method for determining the heading angle according to claim 1, characterized in that, The method for determining the heading angle also includes: Obtain the historical heading angle observation value of the visual target, and determine the angle difference between the heading angle observation value and the historical heading angle observation value; If the angle difference meets the preset stability condition, the heading angle observation value is filtered and corrected according to the confidence level to obtain the corrected heading angle observation value. If the angle difference does not meet the preset stability condition, the historical heading angle observation value of the most recent frame will be used as the heading angle observation value of the current frame.

5. The method for determining the heading angle according to claim 4, characterized in that, If the angle difference does not meet the preset stability condition, then the historical heading angle observation value of the most recent frame will be used as the heading angle observation value of the current frame, including: If the angle difference is not within the preset range, an abnormal count of the heading angle observation value is performed. When the cumulative count exceeds the preset threshold, it is determined that the angle difference does not meet the preset stability condition. The historical heading angle observation value of the corresponding most recent frame is used as the heading angle observation value of the visual target, and the count is set to zero.

6. The method for determining the heading angle according to claim 1, characterized in that, Determining the target heading angle based on the confidence level, the observed heading angle, and the predicted heading angle includes: Based on the confidence level, determine the state update gain between the observed heading angle and the predicted heading angle; Determine the residual between the observed heading angle and the predicted heading angle; The target heading angle is obtained based on the state update gain, the heading angle observation, and the residual.

7. The method for determining the heading angle according to claim 1, characterized in that, The method for determining the heading angle also includes: If the visual target is stationary or moving at low speed, and the number of historical heading angle observations of the visual target reaches a preset number, then the multiple historical heading angle observations are converted into direction vectors to obtain a vector set. Cluster analysis is performed on the vector set to obtain multiple directional clusters, and the centroid direction of the directional cluster with the largest number of samples is determined as the target heading angle.

8. A controller, characterized in that, include: The first memory, on which the computer program is stored; A first processor is configured to execute the computer program in the first memory to implement the method of any one of claims 1-7.

9. A vehicle, characterized in that, Includes the controller as described in claim 8.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1-7.

Images