Method and system for identifying CIPV as car-following target in navigation-based autonomous driving
By filtering and predicting information about surrounding vehicles, a vehicle behavior prediction program is constructed to accurately identify CIPV targets. This solves the problem of unstable identification during the following process of autonomous vehicles, improves the accuracy of identification and the reliability of decision-making, and enhances driving safety.
Patent Information
- Application Number
- PCT/CN2026/087729
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-11-28
- Filing Date
- 2026-03-31
- Publication Date
- 2026-11-19
AI Technical Summary
In existing technologies, autonomous vehicles have difficulty accurately identifying the target vehicle ahead during the following process, resulting in the omission or misselection of CIPV candidate targets, misjudgment of lane change intentions, and inability to provide reliable following decision support.
By acquiring information about surrounding vehicles, a set of important vehicles that qualify as CIPV candidates is selected, a vehicle behavior prediction program is constructed, key parameters are quantitatively predicted, and the lane change intention of vehicles is determined by combining the target selection of adjacent lanes and the current lane, thus achieving accurate identification of CIPV targets.
It improves the accuracy of the CIPV candidate set, enhances the accuracy of following target identification and the reliability of lane change prediction, provides accurate and comprehensive reference for vehicle following decisions, and significantly improves driving safety and the rationality of driving decisions.
Smart Images

Figure CN2026087729_19112026_PF_FP_ABST
Abstract
Description
A method and system for CIPV (Continuous Vehicle Image) recognition in navigation-enabled autonomous driving
[0001] This application claims priority to Chinese Patent Application No. 202511780003.6, filed on November 28, 2025, entitled "A Method and System for Identifying Following Vehicles (CIPV) in Navigation-Assisted Autonomous Driving", the entire contents of which are incorporated herein by reference. Technical Field
[0002] This application relates to the field of autonomous driving prediction algorithm technology, and in particular to a method and system for identifying the closest in-path vehicle (CIPV) in navigation-based autonomous driving. Specifically, it involves taking surrounding vehicle information as input, extracting important vehicles, predicting the behavior, speed, distance from the vehicle in front, and speed of the important vehicles, and finally predicting the vehicle trajectory of the important vehicles in the entire prediction time domain. It also intelligently judges whether the vehicle in front in the adjacent lane intends to change lanes, ensuring high environmental adaptability and driving safety. Background Technology
[0003] As autonomous driving technology becomes more widely used, more and more users will activate the autonomous driving function when driving on the road. When autonomous vehicles are following other vehicles or driving at a constant speed, they need to accurately identify the vehicle in front to deal with the cutting-in and cutting-out behavior of surrounding vehicles.
[0004] In related technologies, the identification of following targets is usually achieved by using perception data fusion and trajectory prediction. First, target screening and trajectory prediction are performed based on road signs and sensor data. Then, the intention of the vehicle to change lanes is judged based on a single trajectory deviation or signal feature, thereby determining the following target.
[0005] However, the above methods are prone to missing or misselecting CIPV candidate targets and misjudging lane change intentions, resulting in unstable CIPV locking and failing to provide reliable support for autonomous driving following decisions. Summary of the Invention
[0006] To achieve the above objectives, this application adopts the following technical solution:
[0007] One aspect of this application provides a method for identifying a following target CIPV in navigation-enabled autonomous driving, the method comprising the following steps:
[0008] Acquire surrounding vehicle information and extract core information about surrounding vehicles; based on preset filtering rules, eliminate invalid interfering targets from surrounding vehicles and select an important set of vehicles that meet CIPV candidate qualifications.
[0009] A vehicle behavior prediction program is built based on a set of historical key vehicles. The selected set of important vehicles is input into the vehicle behavior prediction program to predict vehicle behavior trends and quantitatively predict key parameters; the program outputs a complete state prediction sequence for each candidate vehicle.
[0010] Targets in adjacent lanes and the current lane are screened. The vehicles closest to the current vehicle in the current lane and adjacent lanes are identified as potential CIPV targets. The intent of potential CIPV targets is determined and recorded. If the complete state prediction sequence shows whether the vehicle in front of the current vehicle intends to change lanes, the CIPV target is released. If the complete state prediction sequence shows whether the vehicle closest to the current vehicle in the adjacent lane intends to cut in, the target vehicle is recorded as a new CIPV target.
[0011] In one optional implementation, the process of outputting a complete state prediction sequence for each candidate vehicle includes the following steps:
[0012] Extract the current state variables of each candidate vehicle from the set of important vehicles, including position coordinates, velocity vector, and acceleration vector; at the same time, set the prediction time domain length and time step interval according to the vehicle parameters to form the prediction time axis;
[0013] Using the state transition relationship in the pre-built vehicle behavior prediction program, the current state quantity is used as the initial input to derive the state quantity at each time step in turn; by repeatedly applying the state transition relationship, the state quantities at all time steps in the prediction time domain are generated, forming a state quantity evolution sequence.
[0014] From the generated state variable evolution sequence, extract the key parameters of real-time vehicle speed change curve, dynamic distance from the vehicle, and speed fluctuation of the vehicle itself; merge the key parameters with the state variable sequence and organize them into a complete state prediction sequence arranged in time order.
[0015] In one optional implementation, the process of determining intent includes the following steps:
[0016] Based on the predicted lateral displacement and lateral velocity data provided by the complete state prediction sequence, combined with the lane centerline position, lane width and adjustment coefficient, the intent judgment threshold for each potential CIPV target is calculated; the predicted lateral displacement is compared with the intent judgment threshold, and the vehicle intent classification result is output as changing lanes to the right, keeping lanes, or changing lanes to the left.
[0017] When the output intent classification result is that the vehicle in front of this lane is changing lanes to the right or left, it is determined that the vehicle in front of this lane has the intent to change lanes and cut out; then the CIPV target release mechanism is triggered to clear the current following target association.
[0018] When the output intent classification result is that the nearest vehicle in the adjacent lane is changing lanes to the right or left, and the direction of the lane change is pointing towards the vehicle's own lane, it is determined that the vehicle has the intent to cut in; then the CIPV target update mechanism is triggered, and the nearest vehicle in the adjacent lane is identified as the new following target.
[0019] In one optional implementation, the process of calculating the intent judgment threshold for each potential CIPV target includes the following steps:
[0020] First, extract values from the lane centerline position, lane width, and pre-set adjustment coefficients, and combine them to form a baseline lateral range;
[0021] The obtained baseline lateral range is compared with the predicted lateral displacement provided by each potential CIPV target in the full state prediction sequence. Based on the comparison results, the corresponding upper and lower limits of the lateral tolerance are generated. The output is the lateral displacement threshold range for each target.
[0022] Based on the obtained lateral displacement threshold range and combined with the predicted lateral velocity information, the intent judgment threshold for each target is determined.
[0023] In one optional implementation, the process of determining the intent judgment threshold for each target vehicle includes the following steps:
[0024] The system extracts values from the lane centerline position, lane width, and preset adjustment coefficients, and then combines these values to form a baseline lateral range. The lane centerline position serves as the longitudinal reference benchmark, the lane width serves as the lateral space measurement, the preset adjustment coefficients are used to adjust the sensitivity of the lateral range, and the combined processing calculates a lateral distance range by integrating the values.
[0025] The obtained baseline lateral range is compared with the predicted lateral displacement provided by each potential CIPV target in the full state prediction sequence. Based on the comparison results, the upper and lower limits of the lateral tolerance are generated. The comparison process is based on the boundary of the baseline lateral range to evaluate the degree of deviation of the predicted lateral displacement and determine a lateral displacement threshold interval. The lateral displacement threshold interval defines the allowable range of lateral movement of the vehicle.
[0026] Based on the obtained lateral displacement threshold range, combined with the predicted lateral velocity information provided in the complete state prediction sequence, the intent judgment threshold is determined. The combined processing uses the lateral displacement threshold range as a spatial constraint and the predicted lateral velocity as a dynamic indicator. By evaluating whether the lateral velocity will cause the lateral displacement to exceed the threshold range, a comprehensive intent judgment threshold is generated to ultimately determine the vehicle's lane change intent.
[0027] In one alternative implementation, the process of evaluating whether the lateral velocity would cause the lateral displacement to exceed a threshold range includes the following steps:
[0028] The upper and lower limits of the lateral tolerance within the lateral displacement threshold range are used as the right and left spatial boundaries of the vehicle's lateral movement.
[0029] The predicted lateral velocity information is extracted from the complete state prediction sequence, and the lateral velocity direction attribute and magnitude level are parsed to form the lateral motion dynamic features; the direction attribute is used to indicate the tendency of the vehicle's lateral movement, and the magnitude level is used to indicate the intensity of the lateral movement.
[0030] The lateral motion dynamics are fused and evaluated with the right and left spatial boundaries. When the direction attribute points to the right and the magnitude level indicates that the lateral displacement tends to exceed the right spatial boundary, a threshold for judging the intention to change lanes to the right is generated; when the direction attribute points to the left and the magnitude level indicates that the lateral displacement tends to exceed the left spatial boundary, a threshold for judging the intention to change lanes to the left is generated.
[0031] In one optional implementation, the process of forming lateral motion dynamic features includes the following steps:
[0032] A lateral velocity sequence segment is formed by extracting lateral velocity data for a future preset time domain from the complete state prediction sequence. The lateral velocity sequence segment originates from the complete state prediction sequence output by the vehicle behavior prediction program and contains continuous change data of lateral velocity over a future period.
[0033] The dominant direction of the transverse velocity sequence segment is determined by identifying the main sign tendency of the transverse velocity values in the segment and obtaining the direction attribute. The determination process is based on the dominant proportion of data points in the sequence segment, and the direction attribute is distinguished as pointing to the right or pointing to the left.
[0034] The cumulative impact assessment is based on lateral velocity sequence segments. By measuring the combined displacement effect of all lateral velocity data within the sequence segment on the vehicle's lateral position, the magnitude of the impact is determined. The cumulative impact assessment results are used to characterize the severity of lateral movement and serve as a component of the dynamic characteristics of lateral motion.
[0035] In one alternative implementation, the determination process, based on the dominance ratio of data points within a sequence segment, includes the following steps:
[0036] Each data point in the lateral velocity sequence segment is labeled with a symbol, and a direction symbol is assigned to each lateral velocity value to generate a symbol label sequence.
[0037] The directional weights are calculated for the symbol marker sequence by statistically analyzing the ratio of symbols representing the right direction to those representing the left direction, resulting in a directional weight value. The directional weight value reflects the degree of numerical dominance of symbols representing different directions in the symbol marker sequence.
[0038] The direction weight value is compared with a preset direction determination threshold. When the direction weight value exceeds the direction determination threshold, the direction attribute pointing to the right is output; when the direction weight value is lower than the negative value of the direction determination threshold, the direction attribute pointing to the left is output. The comparison and determination process is based on the relative relationship between the direction weight value and the fixed threshold to finally determine the classification result of the direction attribute.
[0039] In one alternative implementation, the core information of surrounding vehicles includes their location coordinates, driving lane, real-time vehicle speed, longitudinal / lateral distance from the vehicle, and vehicle size.
[0040] Another aspect of this application provides a CIPV (Continuous Interaction Vehicle) recognition system for navigation-assisted autonomous driving, used to implement the CIPV recognition method for navigation-assisted autonomous driving, the CIPV recognition system for navigation-assisted autonomous driving comprising:
[0041] The surrounding vehicle filtering module is used to obtain surrounding vehicle information and extract core information of surrounding vehicles; according to preset filtering rules, invalid interference targets are removed from surrounding vehicles, and an important set of vehicles with CIPV candidate qualifications is selected.
[0042] The vehicle behavior prediction module is used to build a vehicle behavior prediction program based on a historical set of key vehicles. The selected set of important vehicles is input into the vehicle behavior prediction program to predict vehicle behavior trends and quantitatively predict key parameters. The module outputs a complete state prediction sequence for each candidate vehicle. The predicted key parameters include the real-time vehicle speed change curve in the future preset time domain, the dynamic distance to the vehicle in front, and the speed fluctuation of the vehicle in front itself.
[0043] The prediction scheme adjustment module is used to filter targets in adjacent lanes and the current lane. The vehicles closest to the current vehicle in the current lane and adjacent lanes are potential CIPV targets. The intent of potential CIPV targets is judged and recorded. When the complete state prediction sequence shows whether the vehicle in front of the current lane intends to change lanes, if so, the CIPV target is released. When the complete state prediction sequence shows whether the vehicle closest to the adjacent lane intends to cut in, if so, the target vehicle is recorded as a new CIPV target.
[0044] In another aspect of this application, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the computer program, it implements the steps of the navigation-assisted autonomous driving CIPV recognition method as described in any of the above embodiments.
[0045] In another aspect, this application provides a computer-readable storage medium storing a computer program, characterized in that, when executed by a processor, the computer program implements the steps of the navigation-assisted autonomous driving CIPV recognition method as described in any of the above embodiments.
[0046] Another aspect of this application provides a computer program product, characterized in that it includes a computer program, which, when executed by a processor, implements the steps of the navigation-assisted autonomous driving CIPV recognition method as described in any of the above embodiments.
[0047] This application collects core information such as the location coordinates of surrounding vehicles and driving lanes, eliminates invalid interference targets according to preset screening rules, and selects a set of important CIPV candidate vehicles. Based on the historical set of key vehicles, a vehicle behavior prediction program is constructed. The candidate set is input to quantitatively predict key parameters such as the vehicle speed change curve and dynamic spacing within a preset time domain, and outputs a complete state prediction sequence. The final CIPV target is determined by combining the prediction sequence. The trajectory deviation trend, steering signal status, and speed adjustment characteristics of the vehicle in front in the adjacent lane are extracted, and the lane change intention is judged by the prediction program. This embodiment effectively eliminates interference targets and improves the accuracy of the CIPV candidate set by combining core information extraction and screening rules. The prediction program trained with historical data realizes quantitative prediction of key parameters, providing data support for the determination of the following target. Multi-dimensional feature analysis and lane change intention judgment of the vehicle in front in the adjacent lane improve the accuracy of following target identification and the reliability of lane change prediction, providing accurate and comprehensive reference for the vehicle's following decision, and significantly improving driving safety and the rationality of driving decisions. Attached Figure Description
[0048] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:
[0049] Figure 1 is a flowchart illustrating the steps of an embodiment of the CIPV (Continuing Vehicle Target) recognition method for navigation-based autonomous driving according to this application.
[0050] Figure 2 is a schematic diagram of an embodiment of the CIPV (Continuing Vehicle Target) recognition method for navigation-assisted autonomous driving according to this application;
[0051] Figure 3 is a schematic diagram of intent determination in an embodiment of the CIPV (Continuing Vehicle Target) recognition method for navigation-assisted autonomous driving according to this application.
[0052] Figure 4 is a schematic diagram of the steps in this application to select an important set of vehicles that are eligible for CIPV candidate status.
[0053] Figure 5 is a flowchart illustrating the steps involved in outputting the complete state prediction sequence for each candidate vehicle in this application.
[0054] Figure 6 is a flowchart illustrating the steps of this application to determine the intent of a potential CIPV target.
[0055] Figure 7 is a functional module diagram of an embodiment of the CIPV recognition system for navigation-assisted autonomous driving according to this application. Detailed Implementation
[0056] The present application will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.
[0057] The following detailed descriptions are exemplary and intended to provide further detailed explanation of this application. Unless otherwise specified, all technical terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application.
[0058] As is known from common technical knowledge, this application can be implemented through other embodiments that do not depart from its spirit or essential characteristics. Therefore, the disclosed embodiments described above are merely illustrative in all respects and are not the only ones. All modifications within the scope of this application or equivalent to this application are included in this application.
[0059] As shown in Figure 1, this embodiment provides an example of a CIPV (Continuous Vehicle Image) identification method for navigation-enabled autonomous driving. In this embodiment, the CIPV identification method specifically includes the following steps:
[0060] Step S1: Obtain information about surrounding vehicles and extract their core information; based on preset filtering rules, remove invalid interfering targets from the surrounding vehicles and select a set of important vehicles that meet CIPV candidate qualifications.
[0061] The key information about surrounding vehicles includes their location coordinates, driving lane, real-time speed, longitudinal / lateral distance from the vehicle, and vehicle size.
[0062] In some embodiments, the preset filtering rules are three-level filtering rules. The first level is based on the elliptical spatial region of the vehicle speed dynamic adjustment, retaining only vehicles within the region. The second level is based on lane correlation filtering, retaining only vehicles in the current lane and the left and right adjacent lanes. The third level is based on motion trend consistency filtering, retaining only vehicles whose deviation angle from the expected path of the vehicle is less than a preset threshold. The set of vehicles obtained after the three-level filtering is an important set of vehicles with CIPV candidate qualifications.
[0063] Step S2: Construct a vehicle behavior prediction program based on the historical key vehicle set. Input the selected key vehicle set into the vehicle behavior prediction program to predict vehicle behavior trends and quantitatively predict key parameters; output the complete state prediction sequence for each candidate vehicle.
[0064] The key parameters for prediction include the real-time vehicle speed change curve within the preset time domain, the dynamic distance to the vehicle in front, and the speed fluctuation of the vehicle in front itself.
[0065] In some embodiments, the process of generating a complete state prediction sequence is as follows: first, the position coordinates, velocity vector, and acceleration vector of the candidate vehicle are extracted as the current state variables; then, the prediction time domain length and time step interval are set in combination with the vehicle parameters to form a prediction time axis; then, the state transition relationship of the vehicle behavior prediction program is used to recursively generate the state variable evolution sequence of all time steps in the prediction time domain; finally, the state variable evolution sequence is integrated with the above key parameters to form a complete state prediction sequence arranged in chronological order.
[0066] Indicatively, the current state variables of each candidate vehicle are extracted from the set of important vehicles, including position coordinates, velocity vector, and acceleration vector. Simultaneously, the prediction time domain length and time step interval are set according to the vehicle parameters, forming a prediction time axis. Using the state transition relationships in a pre-built vehicle behavior prediction program, the acquired current state variables are used as initial input to derive the state variables for each time step sequentially. By repeatedly applying the state transition relationships, the state variables for all time steps within the prediction time domain are generated, forming a state variable evolution sequence.
[0067] Step S3: Filter targets in adjacent lanes and the current lane. The vehicles closest to the current vehicle in the current lane and adjacent lanes are potential CIPV targets. Determine the intent of potential CIPV targets and record them. When the complete state prediction sequence shows whether the vehicle in front of the current lane intends to change lanes, release the CIPV target if so. When the complete state prediction sequence shows whether the vehicle closest to the adjacent lane intends to cut in, record the target vehicle as a new CIPV target if so.
[0068] In other words, the vehicle closest to the vehicle in the current lane and adjacent lanes is selected as a potential CIPV target, and the intent of the potential CIPV target is determined. Based on the complete state prediction sequence, the current CIPV target is released when the vehicle in front of the current lane has the intention to change lanes and cut out. When the vehicle closest to the vehicle in the adjacent lane has the intention to cut into the vehicle's lane, the vehicle is identified as a new CIPV target.
[0069] The lane change intention judgment is based on the predicted lateral displacement and lateral velocity data in the complete state prediction sequence, combined with the lane centerline position, lane width and preset adjustment coefficient to calculate the intention judgment threshold. The output intention judgment results include three categories: right lane change, lane keeping and left lane change.
[0070] Preferably, in this embodiment, based on the current speed of the vehicle, surrounding vehicles within a certain distance range are marked as important surrounding vehicles, and the current state quantities of the important surrounding vehicles are extracted and identified: x-coordinate, y-coordinate, longitudinal velocity, lateral velocity, longitudinal acceleration, and lateral acceleration. Within one cycle, the acceleration is constant, and based on the current state quantities, the state quantities at different times in the entire future cycle can be deduced.
[0071] Wherein, the state variable X at different times is:
[0072] Wherein, the left column vector represents the vehicle state variables at time k+1, x is the longitudinal position coordinate, y is the longitudinal position coordinate, and v x Let v be the longitudinal velocity vector. y Let a be the transverse velocity vector. x Let a be the longitudinal acceleration vector. y The vector represents the lateral acceleration, the middle matrix is the state transition matrix, Δt is the time step interval, and the column vectors on the right represent the vehicle state variables at time k.
[0073] For the Tp prediction time domain, the state equations at each time point are established as follows:
[0074] Where t0 is the current time; Tp is the prediction time domain, which can take values of 5-10s (depending on vehicle model parameters); each item in the left column vector represents the vehicle state quantity at each time point in the prediction time domain; A is the one-step state transition matrix, A 2 Corresponding to a 2-step state transition, To predict the total number of time-domain steps, the state transition matrix, X t0 This represents the initial state of the vehicle.
[0075] Based on the predicted lateral displacement state quantity y and lateral velocity Vy of surrounding vehicles within a cycle, the vehicle's intention is determined. The determination result is right lane change, lane keeping, or left lane change, represented as {RLC, LK, LLC} respectively; the intention determination is shown in Figure 3.
[0076] Among them, RLC: Change lanes to the right; LK: Lane keeping; LLC: Change lanes to the left; y sl : Centerline; coordinates to the left are positive, coordinates to the right are negative; w l Lane width; a y Lane width adjustment coefficient; v y : Vehicle lateral speed, positive on the left and negative on the right; k: Speed adjustment coefficient.
[0077] First, targets in adjacent lanes and the current lane are screened. The vehicles closest to the current vehicle in the current lane and adjacent lanes are considered potential CIPV targets. A state equation is established in the Tp prediction time domain, and various state variables are calculated. Intent is determined and recorded. If a vehicle ahead in the current lane intends to change lanes, the CIPV target is released. If the closest vehicle in the adjacent lane intends to cut in, it is recorded as a new CIPV target. For details, refer to Figure 2. Step 210: Input perception results; Step 220: Screening of important vehicles (vehicles closest to the current vehicle in the current lane and adjacent lanes); Step 230: Establishing state equations in the TP prediction time domain; Step 240: Lane change intent determination; Step 250: CIPV determination, target identification, and output results.
[0078] In summary, the principle is illustrated in Figure 2. This embodiment collects core information such as the location coordinates of surrounding vehicles and driving lanes, eliminates invalid interference targets according to preset screening rules, and selects a set of important CIPV candidate vehicles. Based on the historical set of key vehicles, a vehicle behavior prediction program is constructed. The candidate set is input to quantitatively predict key parameters such as the vehicle speed change curve and dynamic spacing within a preset time domain, and outputs a complete state prediction sequence. The final CIPV target is determined by combining the prediction sequence. The trajectory deviation trend, steering signal status, and speed adjustment characteristics of the vehicle in front in adjacent lanes are extracted, and the lane change intention is judged by the prediction program. This embodiment effectively eliminates interference targets and improves the accuracy of the CIPV candidate set by combining core information extraction and screening rules. The prediction program trained with historical data realizes quantitative prediction of key parameters, providing data support for determining the following target. Multi-dimensional feature analysis and lane change intention judgment of the vehicle in front in adjacent lanes improve the accuracy of following target identification and the reliability of lane change prediction, providing accurate and comprehensive reference for the vehicle's following decision, and significantly improving driving safety and the rationality of driving decisions.
[0079] Furthermore, as shown in Figure 4, the process of selecting the important set of vehicles with CIPV candidate qualifications in step S1 specifically includes the following steps:
[0080] Step S11: Obtain raw data of all vehicles around the vehicle from the environmental perception system, and extract the core information of each vehicle, including the position coordinates of the longitudinal and lateral coordinates with the vehicle as the reference frame, the driving lane markings, the real-time vehicle speed vector, the longitudinal and lateral distances between the vehicle and the vehicle, and the vehicle size.
[0081] Step S12: Apply the preset three-level filtering rules to eliminate invalid interference targets.
[0082] The first level of filtering is based on dynamic spatial regions: an elliptical dynamic attention region is defined with the vehicle as the center. The major axis of the dynamic attention region extends in the longitudinal direction, and its length is dynamically adjusted by the real-time speed of the vehicle. The minor axis extends in the lateral direction, and its width is determined based on the current lane width and the number of adjacent lanes. Only vehicles located within this dynamic attention region are retained, and vehicles outside the region are excluded.
[0083] The second level of filtering is based on lane correlation: from the vehicles filtered in the first level, only vehicles whose driving lane is the same as or directly adjacent to the left and right lanes are retained; vehicles in other lanes are considered invalid targets and are removed.
[0084] The third level of screening is based on the consistency of motion trends: From the vehicles after the second level of screening, the speed difference vector between each vehicle and the vehicle itself is calculated, and the deviation angle between the vehicle's motion direction and the vehicle's expected path direction is evaluated; only vehicles with a deviation angle less than a preset threshold are retained to ensure that the motion trend of these vehicles is basically consistent with the vehicle's path.
[0085] Step S13: After three levels of screening, the output vehicles constitute an important vehicle set and are qualified as CIPV candidates.
[0086] Preferably, this embodiment retains only vehicles that are highly correlated with the vehicle's driving path in terms of space, lane, and movement trend. This significantly reduces the number of targets the perception system needs to process, improving decision-making focus and response speed. The elliptical area of interest expands and contracts in real time with the vehicle's speed, allowing for coverage of more distant targets at high speeds and narrowing the range at low speeds or in congested conditions, ensuring the filtering range always matches the actual driving risk. Lane association filtering eliminates targets far from the lane that are unrelated to the vehicle's driving direction, avoiding misjudgments caused by cross-lane noise and improving the reliability of target association. Evaluation based on speed difference vectors and deviation angles ensures that the retained vehicles are basically consistent with the vehicle's expected path in terms of movement direction, reducing misidentification caused by lateral or opposite moving targets. During the filtering process, the core features of each target, such as position information, speed vector, and size, are still retained, providing complete and high-quality data support for path planning, collaborative control, and safety decisions. This multi-level, space-, lane-, and motion-based comprehensive filtering mechanism can effectively suppress noise interference and false detections in complex traffic environments, improving the robustness and reliability of the overall perception system.
[0087] Furthermore, the process of evaluating the deviation angle between the vehicle's direction of motion and the expected path direction in step 12 specifically includes the following steps:
[0088] Step S121: Based on the current lane identifier of the vehicle, generate a baseline for the vehicle's expected path; the baseline for the vehicle's expected path is a virtual line parallel to the center line of the lane, and its curvature is determined by the lane's geometric properties; process each candidate vehicle after the second-level screening, and extract the real-time vehicle speed vector containing magnitude and direction for each candidate vehicle; at the same time, obtain the real-time vehicle speed vector of the vehicle.
[0089] Step S122: Combine the speed vector of the candidate vehicle with the speed vector of the self vehicle to obtain a relative motion vector describing the motion trend of the candidate vehicle relative to the self vehicle; project the combined relative motion vector onto the baseline of the self vehicle's expected path, and measure the directional angle difference between the relative motion vector and the baseline of the self vehicle's expected path. The directional angle difference is the deviation angle.
[0090] Step S123: Compare the calculated deviation angle with a preset angle threshold; only candidate vehicles with a deviation angle less than the preset threshold are judged to have a movement trend that is basically consistent with the vehicle path, and thus retained to enter the important vehicle set; vehicles with a deviation angle exceeding the preset threshold are eliminated.
[0091] Preferably, in this embodiment, the expected path baseline is generated based on the vehicle's lane centerline, ensuring that the motion trend assessment always revolves around the vehicle's actual driving route, guaranteeing that the screening results are highly consistent with the vehicle's driving direction. The relative motion vector is synthesized by combining the candidate vehicle's speed vector with the vehicle's speed vector, eliminating differences in vehicle speed magnitude and retaining only the relative motion trend, thus achieving fair comparison of vehicles at different speeds. The relative motion vector is projected onto the expected path baseline, and the difference in direction angle between it and the baseline is measured, obtaining a clear deviation angle value, providing a quantifiable criterion for threshold comparison. By comparing the deviation angle with a preset threshold, only vehicles with directional deviations within the tolerable range are retained, eliminating interfering targets whose motion direction does not match the vehicle's path, significantly reducing the false detection rate and improving the relevance of important vehicle sets. The evaluation process combines real-time vehicle speed vectors and lane geometry attributes, maintaining consistent screening criteria under different road curvatures and traffic scenarios, enhancing the system's robustness in complex road environments. The retained vehicles all possess a motion trend basically consistent with the vehicle's path direction, providing high-quality and reliable target information for path planning, collaborative control, and safety decision-making.
[0092] Furthermore, the process of measuring the difference in direction angle between the relative motion vector and the baseline of the vehicle's expected path in step S122 specifically includes the following steps:
[0093] Step S1221: Based on the generated vehicle expected path baseline, draw a straight line tangent to the vehicle expected path baseline at the current longitudinal position of the vehicle. The direction of the straight line is the instantaneous tangential direction of the vehicle expected path baseline at the current point. Based on the instantaneous tangential direction, construct a path tangential vector of unit length.
[0094] Step S1222: Project the relative motion vector obtained from the path tangential vector synthesis process onto the constructed path tangential vector; the projection operation generates a component along the path tangential direction; at the same time, the difference between the relative motion vector and the tangential component obtained from the projection forms a vertical component perpendicular to the expected path baseline of the vehicle.
[0095] Step S1223: Based on the obtained tangential and vertical components, calculate the ratio of the magnitude of the vertical component to the magnitude of the tangential component; perform arctangent operation on the ratio, and the result is the difference in direction angle between the relative motion vector and the baseline of the vehicle's expected path.
[0096] Preferably, this embodiment projects the relative motion vector onto the path tangential vector to obtain the tangential and vertical components, and then obtains the directional deviation between the two through arctangent operation, achieving accurate numerical representation of the motion trend deviation. Using the instantaneous tangential direction of the vehicle's expected path baseline as a reference, the angle measurement is always consistent with the vehicle's current driving direction, ensuring that the evaluation result is highly correlated with the actual driving path. After separating the tangential and vertical components, only the proportional relationship between the two is considered, effectively filtering out noise components in the relative motion vector that are unrelated to the path direction, improving the robustness of angle estimation. The obtained directional angle difference provides a unified and comparable criterion for subsequent threshold comparison, facilitating a consistent screening standard under different traffic scenarios and vehicle speed conditions. Projection and modulus ratio operations are both basic linear algebra operations, with a simple and efficient calculation process, suitable for the high-speed execution requirements of real-time perception systems.
[0097] Furthermore, the process of forming a vertical component perpendicular to the vehicle's expected path baseline in step S1222 specifically includes the following steps:
[0098] Step S12221: Define the tangential component along the path generated by the projection operation as a vector structure with a clear direction and magnitude; fully preserve the vector properties along the tangential vector direction.
[0099] Step S12222: Perform a vector subtraction operation between the initially obtained original relative motion vector and the tangential component vector structure recorded in the first process to generate a new vector difference.
[0100] Step S12223: Compare the new vector difference with the instantaneous normal direction of the vehicle's expected path baseline. When the two directions are consistent, the vector difference is finally confirmed as the vertical component perpendicular to the vehicle's expected path baseline.
[0101] Preferably, in this embodiment, the projection result along the tangential direction of the path is encapsulated into a vector structure with direction and magnitude, ensuring that subsequent calculations are based on unified and complete vector attributes. The difference obtained by subtracting the tangential component from the original relative motion vector is the component of the relative motion vector in the normal direction of the path baseline, achieving accurate separation of vertical information. The difference vector is compared with the instantaneous normal direction of the expected path baseline of the vehicle, and the difference is confirmed as the vertical component only when the directions are consistent, ensuring that the geometric meaning of the vertical component strictly corresponds to the path baseline. Through explicit tangential / vertical separation and normal comparison, motion components unrelated to the path direction are filtered out, improving the reliability and anti-interference ability of angle deviation calculation.
[0102] Furthermore, as shown in Figure 5, the process of outputting the complete state prediction sequence for each candidate vehicle in step S2 specifically includes the following steps:
[0103] Step S21: Extract the current state variables of each candidate vehicle from the set of important vehicles, including position coordinates, velocity vector, and acceleration vector. Simultaneously, set the prediction time domain length and time step interval based on the vehicle parameters to form a prediction time axis.
[0104] The important vehicle set consists of surrounding vehicles that have been pre-screened, are highly relevant to the vehicle's travel path, and qualify as candidate vehicles to follow. The current state variables are all the basic information used to describe the instantaneous motion characteristics of the vehicles. The position coordinates represent the vehicle's spatial location on the road, the velocity vector represents the vehicle's speed and direction, and the acceleration vector represents the trend and intensity of the vehicle's velocity change. The prediction time domain length refers to the total time range within which the system makes forward predictions of the vehicle's future motion state, used to determine the farthest point in time for prediction. The time step interval refers to the smallest time unit between two adjacent computation nodes during the prediction process. The prediction time axis is a continuous time reference constructed by the prediction time domain length and the time step interval, used to unify the prediction time sequence of all candidate vehicles.
[0105] From the selected set of important vehicles, the driving status information of each candidate vehicle at the current moment is extracted. This information can fully reflect the real-time motion of the vehicle, specifically including its position, speed, and acceleration. Simultaneously, considering the vehicle's driving status, computational performance, and control requirements, a reasonable total prediction duration and minimum time interval are set to construct a continuous and regular prediction timeline, providing a unified time reference for subsequent segmented predictions of vehicle states.
[0106] Step S22: Using the state transition relationship in the pre-built vehicle behavior prediction program, the current state quantity is used as the initial input to derive the state quantity of each time step in turn; by repeatedly applying the state transition relationship, the state quantities of all time steps in the prediction time domain are generated to form a state quantity evolution sequence.
[0107] The vehicle behavior prediction program is a computational program trained on historical vehicle motion data, capable of predicting future vehicle motion trends. State transition relationships are built-in vehicle motion rules used to map the current state to the state at the next moment, ensuring the prediction process conforms to the logic of real vehicle motion. The initial input refers to the current state quantity used to initiate the prediction, serving as the starting point for the entire prediction process. The state quantity at each time step refers to the vehicle motion information corresponding to each smallest time node on the prediction timeline. Repeated application of state transition relationships means recursively applying them sequentially along the timeline, without skipping or omitting any time node. The state quantity evolution sequence is a set of vehicle states arranged chronologically, completely covering the entire prediction time domain, and continuously reflecting the future motion changes of the vehicle.
[0108] By utilizing a vehicle behavior prediction program pre-trained based on historical data, and invoking its preset state transition rules, the program uses the vehicle's current driving state information extracted in the previous step as the starting point for prediction. Following the minimum time interval of the prediction timeline, it progressively calculates the vehicle's driving state for each future time node. By repeating this state calculation process throughout the entire prediction period, continuous vehicle state information covering the entire prediction time period is obtained, forming a complete sequence of vehicle state changes.
[0109] Step S23: Extract key parameters such as real-time vehicle speed change curve, dynamic distance from the vehicle, and vehicle speed fluctuation from the generated state variable evolution sequence; merge the key parameters with the state variable sequence to form a complete state prediction sequence arranged in chronological order.
[0110] The real-time vehicle speed change curve is vehicle speed information extracted from state variables and continuously changing over time, used to reflect vehicle acceleration and deceleration trends; the dynamic distance to the candidate vehicle is the change in the relative distance between the candidate vehicle and the candidate vehicle over time, and is the core basis for following decisions; the vehicle's own speed fluctuation refers to the degree of smooth or drastic change in vehicle speed, used to identify abnormal driving behavior. Key parameters are high-order features extracted from the basic state and directly serve intention judgment and target recognition; the complete state prediction sequence is structured data formed by fusing the basic motion state with high-order key features, with complete temporal sequence and comprehensive information, providing a unified and reliable input for subsequent lane change intention judgment and CIPV target determination.
[0111] From the generated vehicle state change sequence, the most critical information for judging the following target is extracted, including the future speed change trend of the vehicle, the real-time distance change between the vehicle and the following vehicle, and the stability of the vehicle's own speed. These key features are then integrated with the original vehicle state change sequence in chronological order to form a complete and comprehensive vehicle state prediction sequence, providing reliable data support for subsequent judgment of vehicle lane change intentions and determination of the following target.
[0112] Preferably, this embodiment synchronously extracts key state variables such as position, speed, and acceleration from a set of important vehicles, and generates a unified prediction timeline based on the vehicle's parameters to ensure that subsequent predictions are performed on the same time reference. Utilizing the state transition relationships of the vehicle behavior prediction program, the state of each time step is recursively deduced from the current state, achieving multi-step forward prediction of vehicle motion and forming a complete state evolution sequence. From this complete state sequence, key parameters such as vehicle speed change curves, dynamic distance to the vehicle, and speed fluctuations are extracted. These key parameters are then integrated with the full set of state data in chronological order to generate a structured, complete prediction sequence, providing high-quality, time-consistent input for path planning, collaborative control, and safety assessment.
[0113] Furthermore, the process of generating the state variables for all time steps in the prediction time domain in step S22 specifically includes the following steps:
[0114] Step S221: Take the extracted candidate vehicle current state quantity as input, apply the predefined state transition relationship in the vehicle behavior prediction program, and directly output the state quantity of the first time step on the prediction time axis.
[0115] Step S222: Take the state quantity of the first time step as the new input, apply the same state transition relationship again, and output the state quantity of the second time step on the prediction time axis;
[0116] Step S223: Take the output state quantity of the second time step as input, continue to apply the state transition relationship in a loop, and output the state quantities of all subsequent time steps on the prediction time axis in sequence until the set prediction time domain endpoint is covered, and finally form a complete state quantity evolution sequence.
[0117] Preferably, this embodiment achieves continuous evolution of candidate vehicles throughout the entire prediction time domain by recursively applying a unified state transition relationship; each step uses the newly generated state as the input for the next step, ensuring the temporal consistency and dynamic coherence of the state sequence; this process can fully cover all time steps within the set prediction range, providing multi-step forward-looking state information from the current moment, and providing reliable time-series data support for path planning and safety assessment.
[0118] Furthermore, the process of applying the predefined state transition relationships in the vehicle behavior prediction program in step S221 specifically includes the following steps:
[0119] Step S2211: Retrieve the state data records of the target vehicle at multiple consecutive historical moments from the memory unit of the vehicle behavior prediction program, including position coordinates, velocity components and acceleration components, and arrange them in chronological order to form a historical state sequence.
[0120] Step S2212: Perform time series analysis on the formed historical state sequence, identify the change pattern between state quantities at adjacent time moments, establish the mapping relationship between each state quantity at the current time and the corresponding state quantity at the next time moment, and form a state change pattern;
[0121] Step S2213: The extracted state change pattern is parameterized and constructed into a mathematical relation with a fixed structure. The mathematical relation can receive the current state input and output the predicted state value of the next moment, thus completing the standardized encapsulation of the state transition relation.
[0122] Preferably, this embodiment achieves a learnable representation of vehicle behavior by retrieving historical state sequences, analyzing temporal patterns, and parametric modeling; it abstracts the actual motion data at multiple moments into a unified mathematical mapping relationship, enabling the current state to be directly mapped to the predicted state at the next moment; this standardized state transition model not only preserves the dynamic characteristics of vehicle motion but also provides a fast and unified calculation interface, providing a reliable foundation for multi-step prediction.
[0123] Furthermore, the process of constructing a mathematical relation with a fixed structure in step S2213 specifically includes the following steps:
[0124] Step S22131: Based on the association rules between state variables identified in the state change pattern, set an input variable for each current state variable and an output variable for each state variable at the next time step;
[0125] Step S22132: Based on the numerical correspondence patterns revealed by the state change patterns, determine the combination method between each output variable and a specific input variable to form a relational structure framework composed of combinations of input variables;
[0126] Step S22133: Based on the quantified numerical correspondence in the state change pattern, assign specific coefficient values to each combination item in the relational structure framework, multiply the coefficient values with the corresponding variable combinations and then add them together to form a complete mathematical relation.
[0127] Preferably, this embodiment achieves deterministic numerical mapping between vehicle states through systematic modeling of variable mapping, combination structure, and coefficient assignment; it transforms multidimensional inputs into unified algebraic expressions, enabling the direct calculation of the predicted state at the next moment from the current state; the mathematical relation of this fixed structure combines computational efficiency and interpretability, providing a fast and reliable numerical deduction basis for subsequent recursive prediction.
[0128] Furthermore, as shown in Figure 6, the process of determining intent in step S3 specifically includes the following steps:
[0129] Step S31: Based on the predicted lateral displacement and lateral velocity data provided by the complete state prediction sequence, and combined with the lane centerline position, lane width and adjustment coefficient, calculate the intent judgment threshold for each potential CIPV target; compare the predicted lateral displacement with the intent judgment threshold, and output the vehicle intent classification result as changing lanes to the right, keeping lanes, or changing lanes to the left.
[0130] The predicted lateral displacement characterizes the change in a vehicle's lateral position relative to the lane centerline over a future period, while the predicted lateral velocity characterizes the speed and trend of the vehicle's lateral movement. The lane centerline position serves as the reference point for determining the vehicle's lateral position, the lane width is the actual available width of the current lane, and the adjustment coefficient adjusts the sensitivity of intent judgment to avoid misjudgments caused by minor lane adjustments. The intent judgment threshold is the critical standard for distinguishing whether a vehicle has actually changed lanes. By comparing the predicted lateral displacement with this threshold, the vehicle's driving intent can be categorized into three distinct types: changing lanes to the right indicates that the vehicle is about to leave the current lane to the right; maintaining lanes indicates that the vehicle continues to travel within the current lane; and changing lanes to the left indicates that the vehicle is about to leave the current lane to the left.
[0131] Step S32: When the output intent classification result is that the vehicle in front of this lane is changing lanes to the right or left, it is determined that the vehicle in front of this lane has the intent to change lanes and cut out; then the CIPV target release mechanism is triggered to clear the current following target association.
[0132] The vehicle ahead in this lane refers to a vehicle currently in the vehicle's driving lane, directly in front of the vehicle, and locked by the system as a following target. Lane change / cut-out intent means that the vehicle is about to leave the vehicle's lane and is no longer considered a following target. The CIPV target release mechanism is a pre-set target switching logic used to decouple the autonomous driving system from the original following target. After the current following target association is cleared, the system no longer uses that vehicle as a basis for speed and distance control, avoiding safety risks caused by the target continuing to follow after cutting out.
[0133] Step S33: When the output intent classification result is that the nearest vehicle in the adjacent lane is changing lanes to the right or left, and the lane change direction is pointing towards the vehicle's lane, it is determined that the vehicle has the intent to cut in; then the CIPV target update mechanism is triggered, and the nearest vehicle in the adjacent lane is identified as the new following target.
[0134] The nearest vehicle in the adjacent lane refers to the vehicle directly adjacent to the vehicle's lane and closest to it in longitudinal distance; a lane-changing direction pointing towards the vehicle's lane indicates that the vehicle's lane-changing trajectory will enter the vehicle's current lane, constituting a cutting-in behavior that directly affects the vehicle's driving; a cutting-in intention indicates that a vehicle in the adjacent lane is about to enter the vehicle's lane and will become the new vehicle to follow. The CIPV target update mechanism is a pre-set target replacement logic used by the system to prioritize vehicles about to cut in as new following targets, allowing the autonomous driving system to adjust its speed and distance in advance, improving following stability and driving safety.
[0135] Preferably, this embodiment calculates the intent threshold by combining the predicted results of lateral displacement and velocity with lane geometry information to achieve real-time intent classification of potential CIPV targets; based on the classification results, it triggers target release or target update mechanisms respectively, so that when a lane change intention appears ahead of the current lane, the existing following association can be released in time to avoid misfollowing; when an intention to cut into an adjacent lane appears, it can be quickly marked as a new following target to ensure that the following object is always the vehicle most likely to enter the current lane; overall, it forms an intent perception closed loop based on predicted trajectory, improving the accuracy of target selection and the system's response speed to lane change behavior.
[0136] Furthermore, the process of calculating the intent judgment threshold for each potential CIPV target in step S31 specifically includes the following steps:
[0137] Step S311: First, extract values from the lane centerline position, lane width, and preset adjustment coefficient, and combine them to form a reference lateral range;
[0138] Step S312: Compare the obtained baseline lateral range with the predicted lateral displacement provided by each potential CIPV target in the complete state prediction sequence, and generate the corresponding upper and lower limits of lateral tolerance based on the comparison results; the output is the lateral displacement threshold range for each target.
[0139] When the predicted lateral velocity direction of a potential CIPV target deviates from the lane centerline, and its predicted lateral displacement trend points towards the upper boundary of the determined lateral displacement threshold range, the vehicle is determined to have the intention to change lanes to the right. This determination is based on the vehicle moving towards the right edge of the lane at a significant speed. When the predicted lateral velocity direction of a potential CIPV target deviates from the lane centerline, and its predicted lateral displacement trend points towards the lower boundary of the determined lateral displacement threshold range, the vehicle is determined to have the intention to change lanes to the left. This determination is based on the vehicle moving towards the left edge of the lane at a significant speed.
[0140] Step S313: Based on the obtained lateral displacement threshold range and combined with the predicted lateral velocity information, determine the intent judgment threshold for each target vehicle.
[0141] Preferably, this embodiment constructs a baseline lateral range from lane geometry parameters and adjustment coefficients, uses this range to set upper and lower tolerances for predicted lateral displacement and form a threshold interval, and then combines the predicted lateral speed to determine the final intent judgment threshold. This achieves accurate quantification of the lateral behavior boundary of each potential CIPV target, provides a unified, forward-looking and dynamically adaptable criterion for intent classification, and significantly improves the accuracy and robustness of target intent judgment.
[0142] Furthermore, the process of determining the intent judgment threshold for each target vehicle in step S313 specifically includes the following steps:
[0143] Step S3131: Extract values from the lane centerline position, lane width, and preset adjustment coefficients, and form a reference lateral range through value combination processing; wherein, the lane centerline position serves as a longitudinal reference benchmark, the lane width serves as a lateral space measurement, the preset adjustment coefficients are used to adjust the sensitivity of the lateral range, and the combination processing is to comprehensively calculate a lateral distance range by combining the values.
[0144] Step S3132: Compare the obtained baseline lateral range with the predicted lateral displacement provided by each potential CIPV target in the full state prediction sequence, and generate upper and lower limits of lateral tolerance based on the comparison results; the comparison process is based on the boundary of the baseline lateral range to evaluate the degree of deviation of the predicted lateral displacement and determine a lateral displacement threshold range, which defines the allowable range of lateral movement of the vehicle.
[0145] Step S3133: Based on the obtained lateral displacement threshold range and combined with the predicted lateral velocity information provided in the complete state prediction sequence, the intent judgment threshold is determined; the combined processing uses the lateral displacement threshold range as a spatial constraint and the predicted lateral velocity as a dynamic indicator. By evaluating whether the lateral velocity will cause the lateral displacement to exceed the threshold range, a comprehensive intent judgment threshold is generated to ultimately determine the vehicle's lane change intent.
[0146] Preferably, this embodiment constructs a baseline lateral range from lane geometry and adjustment coefficients, uses this range to set upper and lower tolerances for predicted lateral displacement to form a displacement threshold range, and then evaluates this range together with the predicted lateral speed to generate a comprehensive intent judgment threshold. This achieves forward-looking quantification of the lateral behavior boundary of each potential CIPV target, providing a unified, dynamic criterion with spatial-velocity coupling characteristics for lane change intent classification, and significantly improving the accuracy and robustness of intent judgment.
[0147] Furthermore, step S3133, which involves evaluating whether the lateral velocity will cause the lateral displacement to exceed the threshold range, specifically includes the following steps:
[0148] Step S31331: Use the upper limit and lower limit of the lateral tolerance in the lateral displacement threshold range as the right and left spatial boundaries of the vehicle's lateral movement.
[0149] Step S31332: Extract the predicted lateral velocity information from the complete state prediction sequence, parse out the direction attribute and magnitude level of the lateral velocity, and form the lateral motion dynamic features; the direction attribute is used to indicate the tendency of the vehicle's lateral movement, and the magnitude level is used to indicate the intensity of the lateral movement.
[0150] Step S31333: The lateral motion dynamic features are fused and evaluated with the right and left spatial boundaries. When the direction attribute points to the right and the magnitude level indicates that the lateral displacement tends to exceed the right spatial boundary, an intention to change lanes to the right threshold is generated; when the direction attribute points to the left and the magnitude level indicates that the lateral displacement tends to exceed the left spatial boundary, an intention to change lanes to the left threshold is generated.
[0151] Preferably, this embodiment defines the left and right boundaries of the lateral displacement threshold range as spatial constraints, extracts the direction and intensity of the predicted lateral velocity to form motion features, and fuses and evaluates these features with the left and right boundaries. This enables real-time determination of whether a vehicle may break through the left and right spatial constraints within the prediction time domain, thereby generating the corresponding lane change intention threshold and achieving a forward-looking and accurate determination of the lane change intention of potential CIPV targets based on space-velocity coupling.
[0152] Furthermore, the process of forming the lateral motion dynamic features in step S31332 specifically includes the following steps:
[0153] Step S313321: Extract the lateral velocity data of the future preset time domain from the complete state prediction sequence to form a lateral velocity sequence segment; the lateral velocity sequence segment comes from the complete state prediction sequence output by the vehicle behavior prediction program, which contains continuous change data of lateral velocity over a future period of time;
[0154] Step S313322: Determine the dominant direction of the transverse velocity sequence segment, identify the main sign tendency of the transverse velocity values in the segment, and obtain the direction attribute; the determination process is based on the dominant proportion of data points in the sequence segment, and distinguishes the direction attribute as pointing to the right or pointing to the left.
[0155] Step S313323: Perform cumulative impact assessment based on lateral velocity sequence segments. The magnitude of the cumulative impact assessment is obtained by measuring the combined displacement effect of all lateral velocity data within the sequence segment on the vehicle's lateral position. The cumulative impact assessment results are used to characterize the severity of lateral movement and serve as a component of the dynamic characteristics of lateral motion.
[0156] Preferably, this embodiment extracts lateral velocity segments from the complete prediction sequence, determines the dominant direction based on the proportion of positive and negative values in the segments, and then calculates the comprehensive impact of all velocity values in the segments on the lateral displacement and classifies the intensity level, forming a lateral motion dynamic feature that combines directional attributes and intensity level, providing a forward-looking, quantitative criterion with spatial-velocity coupling for intent threshold assessment.
[0157] Furthermore, the determination process in step S313322, which is based on the dominant proportion of data points within a sequence segment, specifically includes the following steps:
[0158] Step S3133221: Perform symbol labeling on each data point in the lateral velocity sequence segment, assign a direction symbol identifier to each lateral velocity value, and generate a symbol label sequence; the symbol label sequence is derived from the mathematical symbol of each data point in the lateral velocity sequence segment;
[0159] Step S3133222: Calculate the directional weights of the symbol marker sequence by statistically analyzing the ratio of symbols representing the right direction to symbols representing the left direction, thus obtaining a directional weight value. The directional weight value reflects the degree of numerical dominance of symbols representing different directions in the symbol marker sequence.
[0160] Step S3133223: Compare and decide the direction weight value with the preset direction determination threshold. When the direction weight value exceeds the direction determination threshold, output the direction attribute pointing to the right; when the direction weight value is lower than the negative value of the direction determination threshold, output the direction attribute pointing to the left. The comparison and decision process determines the classification result of the direction attribute based on the relative relationship between the direction weight value and the fixed threshold.
[0161] Preferably, this embodiment achieves rapid and quantitative determination of the dominant direction of the sequence by marking the transverse velocity segments with symbols, statistically analyzing the proportion of directional symbols and comparing them with a preset threshold, thus forming a clear directional attribute and providing a reliable spatial-velocity direction basis for intent threshold assessment.
[0162] As shown in Figure 7, this embodiment also provides an embodiment of a navigation-based autonomous driving target CIPV recognition system. In this embodiment, the navigation-based autonomous driving target CIPV recognition system is applied to the navigation-based autonomous driving target CIPV recognition method as described in the above embodiments. The navigation-based autonomous driving target CIPV recognition system includes:
[0163] The surrounding vehicle filtering module 1 is used to obtain surrounding vehicle information and extract core information of surrounding vehicles; and with preset filtering rules, remove invalid interference targets from surrounding vehicles and filter out an important set of vehicles that have CIPV candidate qualifications.
[0164] Among them, the core information of surrounding vehicles includes their location coordinates, driving lane, real-time speed, longitudinal / lateral distance from the vehicle, and vehicle size;
[0165] Vehicle behavior prediction module 2 is used to build a vehicle behavior prediction program based on a historical set of key vehicles. The selected set of important vehicles is input into the vehicle behavior prediction program to predict vehicle behavior trends and quantitatively predict key parameters; it outputs a complete state prediction sequence for each candidate vehicle.
[0166] Among them, the key parameters for prediction include the real-time vehicle speed change curve within the future preset time domain, the dynamic distance to the vehicle in front, and the speed fluctuation of the vehicle in front itself.
[0167] The prediction scheme module 3 is used to filter targets in adjacent lanes and the current lane. The vehicles closest to the current vehicle in the current lane and adjacent lanes are potential CIPV targets. The intent of potential CIPV targets is judged and recorded. When the complete state prediction sequence shows whether the vehicle in front of the current lane intends to change lanes, if so, the CIPV target is released. When the complete state prediction sequence shows whether the vehicle closest to the adjacent lane intends to cut in, if so, the target vehicle is recorded as a new CIPV target.
[0168] Preferably, in this embodiment, information about surrounding vehicles is obtained, core information such as location coordinates is extracted, interference is eliminated according to preset rules, and a set of candidate CIPV vehicles is selected; a prediction program is constructed based on historical data, key parameters such as the speed change curve of the candidate set are quantitatively predicted, and a complete state prediction sequence is output; the CIPV is determined by combining the prediction sequence, features such as the trajectory deviation of the vehicle in the adjacent lane are extracted, and the lane change is judged by the prediction program.
[0169] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.
[0170] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more flowchart illustrations and / or one or more block diagrams.
[0171] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams.
[0172] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.
[0173] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and not to limit them. Although this application has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of this application. Any modifications or equivalent substitutions that do not depart from the spirit and scope of this application should be covered within the protection scope of the claims of this application.
Claims
1. A method for identifying a following target CIPV in navigation-assisted autonomous driving, executed by an electronic device, the method comprising the following steps: Obtain information about surrounding vehicles and extract their core information. Based on the preset screening rules, invalid interfering targets are removed from the surrounding vehicles, and an important set of vehicles with CIPV candidate qualifications is selected. A vehicle behavior prediction program is built based on a set of key historical vehicles. The selected set of key vehicles is input into the vehicle behavior prediction program to predict vehicle behavior trends and quantitatively predict key parameters. Output the complete state prediction sequence for each candidate vehicle; Targets in adjacent lanes and the current lane are screened. The vehicles closest to the current vehicle in the current lane and adjacent lanes are identified as potential CIPV targets. The intent of potential CIPV targets is determined and recorded. If the complete state prediction sequence shows whether the vehicle in front of the current vehicle intends to change lanes, the CIPV target is released. If the complete state prediction sequence shows whether the vehicle closest to the current vehicle in the adjacent lane intends to cut in, the target vehicle is recorded as a new CIPV target.
2. The navigation automatic driving follow-up target CIPV recognition method of claim 1, wherein, The process of outputting the complete state prediction sequence for each candidate vehicle includes: Extract the current state variables of each candidate vehicle from the set of important vehicles, including position coordinates, velocity vector, and acceleration vector; at the same time, set the prediction time domain length and time step interval according to the vehicle parameters to form the prediction time axis; Using the state transition relationship in the pre-built vehicle behavior prediction program, the current state quantity is used as the initial input to derive the state quantity at each time step in turn; by repeatedly applying the state transition relationship, the state quantities at all time steps in the prediction time domain are generated, forming a state quantity evolution sequence. From the generated state variable evolution sequence, extract the key parameters of real-time vehicle speed change curve, dynamic distance from the vehicle, and speed fluctuation of the vehicle itself; merge the key parameters with the state variable sequence and organize them into a complete state prediction sequence arranged in time order. 3.The navigation automatic driving follow-up target CIPV identification method of claim 1, wherein, The process of determining the intent of potential CIPV targets includes: Based on the predicted lateral displacement and lateral velocity data provided by the complete state prediction sequence, combined with the lane centerline position, lane width and adjustment coefficient, the intent judgment threshold for each potential CIPV target is calculated; the predicted lateral displacement is compared with the intent judgment threshold, and the vehicle intent classification result is output as changing lanes to the right, keeping lanes, or changing lanes to the left. When the output intent classification result is that the vehicle in front of this lane is changing lanes to the right or left, it is determined that the vehicle in front of this lane has the intent to change lanes and cut out; then the CIPV target release mechanism is triggered to clear the current following target association. When the output intent classification result is that the nearest vehicle in the adjacent lane is changing lanes to the right or left, and the direction of the lane change is pointing towards the vehicle's own lane, it is determined that the vehicle has the intent to cut in; then the CIPV target update mechanism is triggered, and the nearest vehicle in the adjacent lane is identified as the new following target.
4. The navigation automatic driving follow-up target CIPV recognition method of claim 3, wherein, The process of calculating the intent judgment threshold for each potential CIPV target includes: First, extract values from the lane centerline position, lane width, and pre-set adjustment coefficients, and combine them to form a baseline lateral range; The obtained baseline lateral range is compared with the predicted lateral displacement provided by each potential CIPV target in the full state prediction sequence. Based on the comparison results, the corresponding upper and lower limits of the lateral tolerance are generated. The output is the lateral displacement threshold range for each target. Based on the obtained lateral displacement threshold range and combined with the predicted lateral velocity information, the intent judgment threshold for each target is determined.
5. The navigation automatic driving follow-up target CIPV recognition method of claim 4, wherein, The process of determining the intent judgment threshold for each target vehicle includes the following steps: The system extracts values from the lane centerline position, lane width, and preset adjustment coefficients, and then combines these values to form a baseline lateral range. The lane centerline position serves as the longitudinal reference benchmark, the lane width serves as the lateral space measurement, the preset adjustment coefficients are used to adjust the sensitivity of the lateral range, and the combined processing calculates a lateral distance range by integrating the values. The obtained baseline lateral range is compared with the predicted lateral displacement provided by each potential CIPV target in the full state prediction sequence. Based on the comparison results, the upper and lower limits of the lateral tolerance are generated. The comparison process is based on the boundary of the baseline lateral range to evaluate the degree of deviation of the predicted lateral displacement and determine a lateral displacement threshold interval. The lateral displacement threshold interval defines the allowable range of lateral movement of the vehicle. Based on the obtained lateral displacement threshold range, combined with the predicted lateral velocity information provided in the complete state prediction sequence, the intent judgment threshold is determined. The combined processing uses the lateral displacement threshold range as a spatial constraint and the predicted lateral velocity as a dynamic indicator. By evaluating whether the lateral velocity will cause the lateral displacement to exceed the threshold range, a comprehensive intent judgment threshold is generated to ultimately determine the vehicle's lane change intent.
6. The navigation automatic driving follow-up target CIPV recognition method of claim 5, wherein, The process of assessing whether lateral velocity would cause lateral displacement to exceed a threshold range includes: The upper and lower limits of the lateral tolerance within the lateral displacement threshold range are used as the right and left spatial boundaries of the vehicle's lateral movement. The predicted lateral velocity information is extracted from the complete state prediction sequence, and the lateral velocity direction attribute and magnitude level are parsed to form the lateral motion dynamic features; the direction attribute is used to indicate the tendency of the vehicle's lateral movement, and the magnitude level is used to indicate the intensity of the lateral movement. The lateral motion dynamics are fused and evaluated with the right and left spatial boundaries. When the direction attribute points to the right and the magnitude level indicates that the lateral displacement tends to exceed the right spatial boundary, a threshold for judging the intention to change lanes to the right is generated; when the direction attribute points to the left and the magnitude level indicates that the lateral displacement tends to exceed the left spatial boundary, a threshold for judging the intention to change lanes to the left is generated.
7. The navigation automatic driving follow-up target CIPV recognition method of claim 6, wherein, The process of forming lateral motion dynamic characteristics includes: A lateral velocity sequence segment is formed by extracting lateral velocity data for a future preset time domain from the complete state prediction sequence. The lateral velocity sequence segment originates from the complete state prediction sequence output by the vehicle behavior prediction program and contains continuous change data of lateral velocity over a future period. The dominant direction of the transverse velocity sequence segment is determined by identifying the main sign tendency of the transverse velocity values in the segment and obtaining the direction attribute. The determination process is based on the dominant proportion of data points in the sequence segment, and the direction attribute is distinguished as pointing to the right or pointing to the left. The cumulative impact assessment is based on lateral velocity sequence segments. By measuring the combined displacement effect of all lateral velocity data within the sequence segment on the vehicle's lateral position, the magnitude of the impact is determined. The cumulative impact assessment results are used to characterize the severity of lateral movement and serve as a component of the dynamic characteristics of lateral motion.
8. The navigation automatic driving follow-up target CIPV recognition method of claim 7, wherein, The determination process is based on the dominance ratio of data points within a sequence segment, and includes: Each data point in the lateral velocity sequence segment is labeled with a symbol, and a direction symbol is assigned to each lateral velocity value to generate a symbol label sequence. The directional weights are calculated for the symbol marker sequence by statistically analyzing the ratio of symbols representing the right direction to those representing the left direction, resulting in a directional weight value. The directional weight value reflects the degree of numerical dominance of symbols representing different directions in the symbol marker sequence. The direction weight value is compared with a preset direction determination threshold. When the direction weight value exceeds the direction determination threshold, the direction attribute pointing to the right is output; when the direction weight value is lower than the negative value of the direction determination threshold, the direction attribute pointing to the left is output. The comparison and determination process is based on the relative relationship between the direction weight value and the fixed threshold to finally determine the classification result of the direction attribute. 9.The navigation automatic driving follow-up target CIPV identification method of claim 1, wherein, Key information about surrounding vehicles includes their location coordinates, driving lane, real-time speed, longitudinal / lateral distance from the vehicle, and vehicle dimensions.
10. A navigation automatic driving follow target CIPV identification system for implementing the navigation automatic driving follow target CIPV identification method according to any one of claims 1 to 9, characterized in that, The navigation-assisted autonomous driving target CIPV recognition system includes: The surrounding vehicle filtering module is used to obtain surrounding vehicle information and extract core information of surrounding vehicles; according to preset filtering rules, invalid interference targets are removed from surrounding vehicles, and an important set of vehicles with CIPV candidate qualifications is selected. The vehicle behavior prediction module is used to build a vehicle behavior prediction program based on a historical set of key vehicles. The selected set of important vehicles is input into the vehicle behavior prediction program to predict vehicle behavior trends and quantitatively predict key parameters. The module outputs a complete state prediction sequence for each candidate vehicle. The predicted key parameters include the real-time vehicle speed change curve in the future preset time domain, the dynamic distance to the vehicle in front, and the speed fluctuation of the vehicle in front itself. The prediction scheme adjustment module is used to filter targets in adjacent lanes and the current lane. The vehicles closest to the current vehicle in the current lane and adjacent lanes are potential CIPV targets. The intent of potential CIPV targets is judged and recorded. When the complete state prediction sequence shows whether the vehicle in front of the current lane intends to change lanes, if so, the CIPV target is released. When the complete state prediction sequence shows whether the vehicle closest to the adjacent lane intends to cut in, if so, the target vehicle is recorded as a new CIPV target.
11. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the CIPV (Continuous Vehicle Target) identification method for navigation-assisted autonomous driving as described in any one of claims 1 to 9.
12. A computer-readable storage medium, the computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the CIPV identification method for navigation-assisted autonomous driving as described in any one of claims 1 to 9.
13. A computer program product, characterised in that, It includes a computer program that, when executed by a processor, implements the steps of the CIPV (Continuous Vehicle Target) identification method for navigation-assisted autonomous driving as described in any one of claims 1 to 9.
