A vehicle continuous positioning system and method based on unmanned aerial vehicle relay guidance

Abstract

The application discloses a kind of vehicle continuous positioning system and method based on unmanned aerial vehicle relay guidance, comprising: deploying unmanned aerial vehicle relay guidance node, obtaining unmanned aerial vehicle pose information, establishing vehicle communication link;Obtain vehicle motion state information, generate prior state estimation result, send to unmanned aerial vehicle end;Obtain continuous image sequence, input improved LoFTR model, generate visual motion constraint information;Extract multi-path propagation information, identify stable reflection path, construct environment virtual anchor point constraint;Build dynamic confidence solution structure, combine multi-source observation, iteratively update vehicle position state;Carry out consistency determination, execute observation reconstruction operation, maintain vehicle continuous stable positioning output.The application realizes the continuous stable positioning of vehicle in complex environment by fusing unmanned aerial vehicle visual observation, multi-path reflection information and dynamic confidence distribution mechanism.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation and unmanned systems integration technology, and in particular to a vehicle continuous positioning system and method based on unmanned aerial vehicle (UAV) relay guidance. Background Technology

[0002] With the development of intelligent transportation systems and unmanned systems technologies, high-precision vehicle positioning has become a crucial foundation for applications such as autonomous driving, vehicle-to-everything (V2X) communication, and intelligent monitoring. Existing vehicle positioning methods primarily rely on fusion positioning using Global Navigation Satellite Systems (GNSS) combined with inertial navigation, wheel odometers, and other multi-source information, achieving high accuracy in open environments. However, in complex environments such as urban canyons, tunnels, under bridges, and densely populated areas with tall buildings, GNSS signals are easily obstructed, interfered with by multipath effects, or completely fail, leading to a significant decrease in vehicle positioning accuracy or even positioning interruptions. Therefore, how to achieve continuous and stable vehicle positioning under conditions of weak observation or the absence of GNSS has become a key research focus.

[0003] To improve positioning capabilities in complex environments, existing technologies have introduced visual positioning, wireless signal-assisted positioning, and multi-source fusion methods. For example, visual-based positioning methods acquire environmental features through cameras for matching or reconstruction, but feature matching stability is poor under conditions of changing illumination, changing viewpoints, and weak textures. Wireless signal-based positioning methods utilize signal propagation characteristics for positioning, but multipath effects are complex and difficult to model, leading to unstable positioning results. While multi-source fusion methods can improve robustness to some extent, they lack effective dynamic weight allocation and constraint reconstruction mechanisms when observation information is inconsistent or some observations fail, making them prone to error accumulation and positioning drift.

[0004] Existing technologies typically rely on single sensors or fixed fusion strategies, lacking the ability to model the stability and dynamically schedule multi-source observation information over continuous time. This is especially true in scenarios involving UAV-assisted observation, where an effective relay guidance and constraint construction mechanism has not yet been established, making it impossible to fully utilize UAV perspective information and environmental reflection information to construct stable positioning constraints. Consequently, achieving continuous, stable, and highly reliable vehicle positioning results remains challenging in complex environments.

[0005] Therefore, how to provide a vehicle continuous positioning system and method based on UAV relay guidance is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a vehicle continuous positioning system and method based on UAV relay guidance. This invention fully utilizes UAV aerial perspective perception, multi-path wireless propagation information, and an improved LoFTR model to fuse multi-source observation information of vehicles in complex environments, constructs virtual environmental anchor point constraints, and introduces a dynamic credibility constraint allocation mechanism to realize state solving and observation reconstruction in the continuous vehicle positioning process. It has the advantages of strong positioning continuity, high anti-occlusion capability, and stable and reliable positioning results.

[0007] A vehicle continuous positioning method based on UAV relay guidance according to an embodiment of the present invention includes: Deploy drone relay guidance nodes to obtain drone pose information and establish wireless communication links between drones and vehicles; The system acquires the vehicle's motion state information, generates a priori state estimate of the vehicle's current position based on the motion state information, and sends it to the UAV relay guidance node via a wireless communication link. A continuous image sequence containing the vehicle target is obtained. Images at adjacent time points and corresponding UAV pose information are input into the improved LoFTR model to generate visual motion constraint information of the vehicle target. The visual observation of the vehicle relative to the UAV is constructed by combining the UAV pose information. Acquire multipath propagation information of wireless signals between UAV and vehicle, identify stable reflection propagation paths in continuous time and determine environmental reflection positions, and use environmental reflection positions as virtual environmental anchor points to construct vehicle positioning constraints. A state-solving module based on dynamic credibility constraint allocation is constructed to jointly process the vehicle prior state estimation results, visual observations, and vehicle positioning constraints. According to the stability of each observation source in continuous time, a corresponding credibility weight is assigned to each observation source, and the vehicle position state is iteratively solved to obtain the continuous vehicle positioning results. Consistency determination is performed on the continuous vehicle positioning results. When the consistency determination results meet the preset abnormal conditions, the observation reconstruction operation is performed. By adjusting the confidence weight or re-identifying the virtual anchor points of the environment, the continuous vehicle positioning output is maintained.

[0008] Optionally, obtaining the UAV pose information includes: The drone is controlled to fly to a preset airspace above the vehicle or above the vehicle's driving path, and satellite navigation positioning data, inertial measurement data, and barometric altitude data of the drone are acquired. The satellite navigation positioning data, inertial measurement data, and barometric altitude data are time-aligned, and short-term compensation is performed on the satellite navigation positioning data based on the inertial measurement data to obtain the drone's position coordinates and attitude angle information, including roll angle, pitch angle, and heading angle.

[0009] Optionally, establishing a wireless communication link between the drone and the vehicle includes: By scanning signals and initializing links between the UAV communication unit and the vehicle communication unit, a two-way communication connection is established between the UAV and the vehicle. Time synchronization is performed between the two communicating parties. After the communication is established, the link signal strength and communication delay information are continuously acquired. The communication link status is determined based on the signal strength and communication delay information. When the link status meets the preset conditions, the stable connection of the wireless communication link is maintained.

[0010] Optionally, the step of generating a priori state estimation result of the vehicle's current position based on motion state information and sending it to the UAV relay guidance node via a wireless communication link includes: The motion state information of the vehicle at the current moment is obtained. The motion state information includes the vehicle's inertial measurement data, wheel speed data, and steering data. The inertial measurement data, wheel speed data, and steering data are time-aligned according to a unified sampling time sequence to form the vehicle motion state sequence corresponding to the current moment. Based on the vehicle motion state sequence, extract the longitudinal motion change features, lateral motion change features, and heading change features of the vehicle in continuous time moments, and construct the vehicle short-time trajectory evolution sequence based on the longitudinal motion change features, lateral motion change features, and heading change features; The short-term trajectory evolution sequence of the vehicle is subjected to steering-sensitive discrimination processing. When the vehicle is detected to be turning, the weight of the steering data on the trajectory evolution sequence is increased. When the vehicle is detected to be traveling straight, the weight of the wheel speed data on the trajectory evolution sequence is increased, and the direction correction result corresponding to the current driving state is generated. The trajectory evolution sequence and the direction correction result are subjected to continuity constraint processing to suppress position jumps caused by instantaneous fluctuations in continuous time intervals, and the prior trajectory position at the current time is generated according to the motion continuity relationship of the vehicle in continuous time intervals. The prior trajectory position is combined with the current velocity and heading states to generate a prior state estimate of the vehicle's current position, and the prior state estimate is sent to the UAV relay guidance node.

[0011] Optionally, generating visual motion constraint information of the vehicle target and constructing a visual observation of the vehicle relative to the UAV by combining the UAV pose information includes: An improved LoFTR model is constructed, which includes a pose-guided input unit, a target region enhancement unit, a cross-temporal feature preservation unit, a visual motion constraint generation unit, and a credibility feedback correction unit. The pose guidance input unit receives images from adjacent time points, UAV pose information, and vehicle prior state estimation results. Based on the UAV pose information and vehicle prior state estimation results, it determines the predicted region of the vehicle target in the current image and generates corresponding region guidance information. The target region enhancement unit receives image features and region guidance information, performs feature enhancement on the region corresponding to the vehicle target, performs feature suppression on the non-target region, and outputs a target enhancement feature map. Position encoding is performed on the target enhanced feature map. Intra-image feature association and cross-image feature association are performed on the encoded feature map to generate cross-time-time association features. The initial matching result is determined based on the cross-time-time association features. The initial matching result is locally refined to obtain candidate matching features between the current time and the previous time. The cross-temporal feature preservation unit receives candidate matching features and matching trajectory information from continuous historical moments. It jointly determines the position changes, direction changes, and displacement continuity of each matching feature in continuous moments, retains stable matching features that meet the continuity condition, and forms a temporal stable feature set of the vehicle target. The visual motion constraint generation unit receives a time-stable feature set, extracts the displacement change trend, direction change trend and viewpoint change relationship of the vehicle target in continuous time, generates visual motion constraint information of the vehicle target, and forms a relative motion description result of the vehicle target relative to the UAV based on the visual motion constraint information. The credibility feedback correction unit receives the relative motion description results, UAV pose information, and vehicle continuous positioning results at the current moment, performs credibility correction on the visual motion constraint information, and constructs the visual observation of the vehicle relative to the UAV based on the corrected visual motion constraint information and UAV pose information. The improved LoFTR model was trained, and the visual motion constraint reconstruction error was used as the optimization target. The parameters of the pose-guided input unit, target region enhancement unit, cross-temporal feature preservation unit, visual motion constraint generation unit, and credibility feedback correction unit were continuously optimized.

[0012] Optionally, the step of constructing vehicle positioning constraints by using environmental reflection locations as virtual environmental anchor points includes: The system acquires multipath information of wireless propagation signals between the UAV and the vehicle at continuous time intervals, extracts the propagation delay change information and signal strength change information of each propagation path at continuous time intervals, and forms a multipath time sequence information according to the corresponding relationship of the changes of the propagation path at continuous time intervals. Based on the multipath timing information sequence, propagation paths with continuous propagation delay changes and stable signal strength changes in consecutive time moments are screened to determine stable reflection propagation paths, and the corresponding environmental reflection positions are determined based on the propagation characteristics of stable reflection propagation paths in consecutive time moments. By jointly associating the environmental reflection positions with the current pose information of the UAV and the prior state estimation results of the vehicle, the spatial distribution relationship of each environmental reflection position relative to the vehicle's driving direction is determined, and environmental reflection positions located in the current forward region or lateral continuous region of the vehicle's current driving direction are screened to form a candidate set of virtual environmental anchor points. The environmental reflection positions in the candidate set of virtual environmental anchor points are continuously calibrated. Based on the degree of constraint offset of the vehicle prior state estimation result at the same environmental reflection position in consecutive time moments, environmental reflection positions with continuous constraint offset and consistent direction are retained as valid virtual environmental anchor points. Establish corresponding constraints between the effective virtual anchor points of the environment and the prior state estimation results of the vehicle's current position, so that each effective virtual anchor point of the environment forms an independent position convergence constraint on the vehicle's current position. The order of constraint action is determined according to the stability of each effective virtual anchor point in continuous time, and vehicle positioning constraints for continuous vehicle positioning are generated.

[0013] Optionally, the step of assigning corresponding confidence weights to each observation source based on the stability of each observation source in continuous time intervals and iteratively solving the vehicle position state to obtain the continuous vehicle positioning result includes: A state-solving module based on dynamic credibility constraint allocation is constructed. The state-solving module includes a multi-source credibility evolution unit, a constraint priority scheduling unit, and a progressive state convergence unit. The multi-source credibility evolution unit receives the vehicle's prior state estimation results, visual observations, and vehicle positioning constraints, and forms corresponding prior state inputs, visual constraint inputs, and virtual anchor point constraint inputs, respectively. The multi-source credibility evolution unit jointly determines the continuity, directional consistency, and constraint offset stability of the changes in prior state input, visual constraint input, and virtual anchor point constraint input in continuous time intervals, generates dynamic credibility results corresponding to each input, and forms a credibility evolution sequence for each input according to the smoothness of credibility changes in continuous time intervals. The constraint priority scheduling unit receives the confidence evolution sequence and sorts the dynamic confidence results of the prior state input, visual constraint input, and virtual anchor point constraint input in descending order to determine the constraint action order of each input. Then, the corresponding inputs are selected in sequence according to the constraint action order to participate in the solution at the current position, forming the hierarchical constraint combination result at the current time. The progressive state convergence unit receives the results of the hierarchical constraint combination. It first uses the input with the highest priority to generate the initial solution result of the vehicle's current position, and then introduces the remaining inputs in sequence to correct the initial solution result layer by layer. This allows the vehicle's position state, velocity state, and heading state to gradually converge during the continuous correction process, thus obtaining the vehicle state result at the current moment. The vehicle state results are fed back to the multi-source credibility evolution unit to update the dynamic credibility results and constraint action order of each input at the next moment. The vehicle position state, velocity state and heading state after convergence at the current moment are output as the vehicle continuous positioning results.

[0014] Optionally, the observation reconstruction operation, which maintains continuous vehicle positioning output by adjusting confidence weights or re-identifying virtual environmental anchor points, includes: Based on the vehicle's continuous positioning results, the vehicle's prior state estimation results, visual observations, and vehicle positioning constraints, the current position deviation, observation deviation, and constraint deviation are determined to form the consistency judgment result at the current moment. The consistency judgment results in consecutive time moments are compared in time sequence to determine the change state of the current position deviation, observation deviation and constraint deviation in consecutive time moments, and the vehicle continuous positioning result at the current time moment is judged to meet the preset abnormal conditions based on the change state. When the vehicle's continuous positioning results at the current moment meet the preset abnormal conditions, an observation reconstruction operation is performed to redistribute the dynamic credibility results corresponding to each input, redetermine the constraint action order of each input based on the redistributed dynamic credibility results, re-identify and filter the virtual anchor points of the environment, and update the vehicle positioning constraints. Based on the reallocated dynamic reliability results, the redefined constraint order, and the updated vehicle positioning constraints, the solution process for the vehicle position state, speed state, and heading state is re-executed to obtain and output the corrected continuous vehicle positioning results.

[0015] A vehicle continuous positioning system based on unmanned aerial vehicle (UAV) relay guidance according to an embodiment of the present invention includes: The drone relay guidance module is used to acquire drone pose information and establish a wireless communication link between the drone and the vehicle. The vehicle prior state module is used to acquire vehicle motion state information and generate prior state estimation results for the vehicle's current position. The visual observation generation module is used to generate visual motion constraint information of vehicle targets and construct visual observations based on the improved LoFTR model. The virtual anchor point construction module is used to determine the location of environmental reflections and construct vehicle positioning constraints; The state solution module is used to jointly process the vehicle's prior state estimation results, visual observations, and vehicle positioning constraints to perform iterative solution of the vehicle's position state and obtain continuous vehicle positioning results. The observation reconstruction module is used to determine the consistency of continuous vehicle positioning results and to perform observation reconstruction under abnormal conditions to maintain continuous positioning output.

[0016] The beneficial effects of this invention are: This invention introduces a UAV relay guidance mechanism to provide vehicles with a stable aerial observation source in complex environments, effectively compensating for the problem of traditional satellite navigation failing in obstructed environments. By utilizing continuous image information acquired by UAVs and combining it with an improved visual feature association method to generate visual motion constraints for the vehicle, the vehicle can still obtain reliable external observation information even in the absence of satellite signals, thus improving the continuity of positioning in complex environments.

[0017] This invention analyzes the multipath propagation information of wireless signals, extracts stable reflection paths, and constructs virtual anchor point constraints in the environment, transforming previously unusable reflected signals into effective constraint information for positioning. This method breaks through the traditional approach of treating multipath signals merely as interference, achieving proactive utilization of environmental information. Furthermore, by combining a continuous temporal filtering mechanism, it improves the stability and effectiveness of constraints, enhancing the robustness of the overall positioning system.

[0018] This invention constructs a state solution and observation reconstruction mechanism based on dynamic reliability allocation. It dynamically adjusts the effect of multi-source observations according to their stability in continuous time, and performs observation reconstruction and constraint update when the positioning results are abnormal. This effectively suppresses error accumulation and positioning drift problems, enabling the invention to achieve high continuity, high stability and high reliability in complex environments. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a vehicle continuous positioning method based on UAV relay guidance proposed in this invention; Figure 2 This is a schematic diagram of a vehicle continuous positioning system based on UAV relay guidance proposed in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figure 1 A method for continuous vehicle positioning based on UAV relay guidance includes: Deploy drone relay guidance nodes to obtain drone pose information and establish wireless communication links between drones and vehicles; The system acquires the vehicle's motion state information, generates a priori state estimate of the vehicle's current position based on the motion state information, and sends it to the UAV relay guidance node via a wireless communication link. A continuous image sequence containing the vehicle target is obtained. Images at adjacent time points and corresponding UAV pose information are input into the improved LoFTR model to generate visual motion constraint information of the vehicle target. The visual observation of the vehicle relative to the UAV is constructed by combining the UAV pose information. Acquire multipath propagation information of wireless signals between UAV and vehicle, identify stable reflection propagation paths in continuous time and determine environmental reflection positions, and use environmental reflection positions as virtual environmental anchor points to construct vehicle positioning constraints. A state-solving module based on dynamic credibility constraint allocation is constructed to jointly process the vehicle prior state estimation results, visual observations, and vehicle positioning constraints. According to the stability of each observation source in continuous time, a corresponding credibility weight is assigned to each observation source, and the vehicle position state is iteratively solved to obtain the continuous vehicle positioning results. Consistency determination is performed on the continuous vehicle positioning results. When the consistency determination results meet the preset abnormal conditions, the observation reconstruction operation is performed. By adjusting the confidence weight or re-identifying the virtual anchor points of the environment, the continuous vehicle positioning output is maintained.

[0022] In this embodiment, obtaining the UAV pose information includes: The drone is controlled to fly to a preset airspace above the vehicle or above the vehicle's driving path, and satellite navigation positioning data, inertial measurement data, and barometric altitude data of the drone are acquired. The satellite navigation positioning data, inertial measurement data, and barometric altitude data are time-aligned, and short-term compensation is performed on the satellite navigation positioning data based on the inertial measurement data to obtain the drone's position coordinates and attitude angle information, including roll angle, pitch angle, and heading angle.

[0023] In this embodiment, establishing a wireless communication link between the drone and the vehicle includes: By scanning signals and initializing links between the UAV and vehicle communication units, a two-way communication connection is established between the UAV and the vehicle. Time synchronization is performed between the two communicating parties. After the communication is established, the link signal strength and communication delay information are continuously acquired. The communication link status is determined based on the signal strength and communication delay information. When the link status meets the preset conditions, the stable connection of the wireless communication link is maintained. The preset conditions are that the link signal strength is not lower than -75 dBmW and the communication delay is not higher than 50 milliseconds.

[0024] In this embodiment, the step of generating a priori state estimation result of the vehicle's current position based on motion state information and sending it to the UAV relay guidance node via a wireless communication link includes: The motion state information of the vehicle at the current moment is obtained. The motion state information includes the vehicle's inertial measurement data, wheel speed data, and steering data. The inertial measurement data, wheel speed data, and steering data are time-aligned according to a unified sampling time sequence to form the vehicle motion state sequence corresponding to the current moment. Based on the vehicle motion state sequence, extract the longitudinal motion change features, lateral motion change features, and heading change features of the vehicle in continuous time moments, and construct the vehicle short-time trajectory evolution sequence based on the longitudinal motion change features, lateral motion change features, and heading change features; The short-term trajectory evolution sequence of the vehicle is subjected to steering-sensitive discrimination processing. When the vehicle is detected to be turning, the weight of the steering data on the trajectory evolution sequence is increased. When the vehicle is detected to be traveling straight, the weight of the wheel speed data on the trajectory evolution sequence is increased, and the direction correction result corresponding to the current driving state is generated. The trajectory evolution sequence and the direction correction result are subjected to continuity constraint processing to suppress position jumps caused by instantaneous fluctuations in consecutive time steps, and the prior trajectory position at the current time is generated according to the motion continuity relationship of the vehicle in consecutive time steps. Specifically, the continuity constraint processing of the trajectory evolution sequence and the direction correction result is as follows: Consistency detection is performed on the position change amplitude, direction change amplitude, and velocity change amplitude between adjacent trajectory points in continuous time. Abnormal trajectory points with change amplitude exceeding a preset threshold are removed. The retained trajectory points are smoothly connected in chronological order. The trajectory at the current time is corrected based on the trajectory change trend of the previous time to obtain a trajectory sequence that meets the continuous change condition. The prior trajectory position is combined with the current velocity and heading states to generate a prior state estimate of the vehicle's current position, and the prior state estimate is sent to the UAV relay guidance node.

[0025] In this embodiment, generating visual motion constraint information for the vehicle target and constructing a visual observation of the vehicle relative to the UAV by combining the UAV pose information includes: An improved LoFTR model is constructed, which includes a pose-guided input unit, a target region enhancement unit, a cross-temporal feature preservation unit, a visual motion constraint generation unit, and a credibility feedback correction unit. The pose guidance input unit receives images from adjacent time points, UAV pose information, and vehicle prior state estimation results. Based on the UAV pose information and vehicle prior state estimation results, it determines the predicted region of the vehicle target in the current image and generates corresponding region guidance information. The target region enhancement unit receives image features and region guidance information, performs feature enhancement on the region corresponding to the vehicle target, performs feature suppression on the non-target region, and outputs a target enhancement feature map. Positional encoding is performed on the target enhanced feature map. Intra-image feature association and cross-image feature association are then performed on the encoded feature map to generate cross-time-step association features. Initial matching results are determined based on the cross-time-step association features. The initial matching results are then locally refined to obtain candidate matching features between the current time step and the previous time step. Specifically, the generation of cross-time-step association features involves: The encoded feature maps are subjected to intra-image feature self-association processing to extract the context information within each image. The feature maps of adjacent time frames are subjected to cross-image feature interaction processing to establish feature correspondence between different time frames. The feature correspondence is filtered by global consistency constraints, and feature pairs that satisfy the correspondence in both spatial location and feature expression are retained to form cross-time-linked features. The initial matching results are determined based on cross-time correlation features, and then the initial matching results are locally refined, specifically as follows: Based on the correspondence between features in the cross-time correlation features, an initial matching feature pair is determined. Then, with the initial matching feature pair as the center, a local range search is performed in the corresponding image region. The matching position is gradually refined and adjusted to obtain matching features with higher positional accuracy, forming candidate matching features between the current time and the previous time. The cross-temporal feature preservation unit receives candidate matching features and matching trajectory information from continuous historical moments. It jointly determines the position changes, direction changes, and displacement continuity of each matching feature in continuous moments, retains stable matching features that meet the continuity condition, and forms a temporal stable feature set of the vehicle target. The visual motion constraint generation unit receives a time-stable feature set, extracts the displacement change trend, orientation change trend, and viewpoint change relationship of the vehicle target in continuous time steps, generates visual motion constraint information of the vehicle target, and forms a relative motion description result of the vehicle target relative to the UAV based on the visual motion constraint information. Specifically, the visual motion constraint information of the vehicle target is generated as follows: Statistical analysis is performed on the position changes of each feature point in the temporally stable feature set over consecutive time intervals to extract the overall displacement change trend. The main motion direction of the vehicle is determined based on the consistency of the motion direction of each feature point. The viewpoint change is determined by combining the distribution change relationship of the feature points in the image. The displacement change trend, main motion direction and viewpoint change are jointly expressed to form the visual motion constraint information of the vehicle target. The visual motion constraint information includes the displacement change information, motion direction information and viewpoint change information of the vehicle over consecutive time intervals. The credibility feedback correction unit receives the relative motion description results, UAV pose information, and vehicle continuous positioning results at the current moment, performs credibility correction on the visual motion constraint information, and constructs the visual observation of the vehicle relative to the UAV based on the corrected visual motion constraint information and UAV pose information. The improved LoFTR model was trained, and the visual motion constraint reconstruction error was used as the optimization target. The parameters of the pose-guided input unit, target region enhancement unit, cross-temporal feature preservation unit, visual motion constraint generation unit, and credibility feedback correction unit were continuously optimized.

[0026] In this embodiment, the step of using the environmental reflection location as a virtual environmental anchor point to construct vehicle positioning constraints includes: The system acquires multipath information of wireless propagation signals between the UAV and the vehicle at continuous time intervals, extracts the propagation delay change information and signal strength change information of each propagation path at continuous time intervals, and forms a multipath time sequence information according to the corresponding relationship of the changes of the propagation path at continuous time intervals. Based on the multipath timing information sequence, propagation paths with continuous propagation delay changes and stable signal strength changes in consecutive time intervals are screened to determine stable reflection propagation paths. Then, based on the propagation characteristics of the stable reflection propagation paths in consecutive time intervals, the corresponding environmental reflection positions are determined. Specifically, determining the stable reflection propagation path involves: The continuity of the propagation delay changes of each propagation path in the multi-path time sequence is determined, and the stability of the corresponding signal strength changes is determined. The propagation path that simultaneously satisfies the conditions of continuous propagation delay changes and signal strength changes not exceeding a preset range is selected as a stable reflection propagation path. By jointly associating the environmental reflection positions with the current pose information of the UAV and the prior state estimation results of the vehicle, the spatial distribution relationship of each environmental reflection position relative to the vehicle's driving direction is determined, and environmental reflection positions located in the current forward region or lateral continuous region of the vehicle's current driving direction are screened to form a candidate set of virtual environmental anchor points. The environmental reflection positions in the candidate set of virtual environmental anchor points are continuously calibrated. Based on the degree of constraint offset of the same environmental reflection position to the vehicle prior state estimation result in continuous time, the environmental reflection positions with continuous constraint offset and consistent direction are retained as effective virtual environmental anchor points. The degree of constraint offset refers to the change in positional difference between the vehicle positioning constraint formed based on the same environmental reflection position and the vehicle prior state estimation result in continuous time. Establish corresponding constraints between the effective virtual anchor points of the environment and the prior state estimation results of the vehicle's current position, so that each effective virtual anchor point of the environment forms an independent position convergence constraint on the vehicle's current position. The order of constraint action is determined according to the stability of each effective virtual anchor point in continuous time, and vehicle positioning constraints for continuous vehicle positioning are generated.

[0027] In this embodiment, the step of assigning corresponding confidence weights to each observation source based on the stability of each observation source over consecutive time intervals and iteratively solving the vehicle position state to obtain the continuous vehicle positioning result includes: A state-solving module based on dynamic credibility constraint allocation is constructed. The state-solving module includes a multi-source credibility evolution unit, a constraint priority scheduling unit, and a progressive state convergence unit. The multi-source credibility evolution unit receives the vehicle's prior state estimation results, visual observations, and vehicle positioning constraints, and forms corresponding prior state inputs, visual constraint inputs, and virtual anchor point constraint inputs, respectively. The multi-source credibility evolution unit jointly determines the continuity, directional consistency, and constraint offset stability of the changes in prior state input, visual constraint input, and virtual anchor point constraint input over consecutive time steps, generating dynamic credibility results corresponding to each input. Furthermore, it forms a credibility evolution sequence for each input based on the smoothness of credibility changes over consecutive time steps. Specifically, the generation of dynamic credibility results corresponding to each input is as follows: The continuity of positional changes of prior state input, visual constraint input, and virtual anchor point constraint input in consecutive time steps is determined to obtain a continuity evaluation result. The consistency of directional changes of each input in consecutive time steps is determined to obtain a directional consistency evaluation result. The stability of offset changes of each input to the vehicle positioning result in consecutive time steps is determined to obtain a constraint offset stability evaluation result. The continuity evaluation result, directional consistency evaluation result, and constraint offset stability evaluation result are comprehensively processed to generate the dynamic credibility result corresponding to each input. The constraint priority scheduling unit receives the confidence evolution sequence and sorts the dynamic confidence results of the prior state input, visual constraint input, and virtual anchor point constraint input in descending order of their respective values ​​at the current time. This determines the constraint action order of each input and selects the corresponding inputs to participate in the solution at the current position in the order of constraint action, forming the hierarchical constraint combination result at the current time. The constraint action order of each input is as follows: prior state input, visual constraint input, and virtual anchor point constraint input are arranged in descending order of their respective dynamic confidence results. The input with the largest dynamic confidence result value has the highest constraint action order, and the input with the smallest dynamic confidence result value has the lowest constraint action order. The progressive state convergence unit receives the results of the hierarchical constraint combination. It first uses the input with the highest priority to generate the initial solution result of the vehicle's current position, and then introduces the remaining inputs in sequence to correct the initial solution result layer by layer. This allows the vehicle's position state, velocity state, and heading state to gradually converge during the continuous correction process, thus obtaining the vehicle state result at the current moment. The vehicle state results are fed back to the multi-source credibility evolution unit to update the dynamic credibility results and constraint action order of each input at the next moment. The vehicle position state, velocity state and heading state after convergence at the current moment are output as the vehicle continuous positioning results.

[0028] In this embodiment, the observation reconstruction operation, which maintains continuous vehicle positioning output by adjusting the confidence weight or re-identifying virtual environmental anchor points, includes: Based on the vehicle's continuous positioning results, prior state estimation results, visual observations, and vehicle positioning constraints, the current position deviation, observation deviation, and constraint deviation are determined to form the consistency judgment result for the current moment, wherein: The position deviation is determined by comparing the difference between the vehicle's continuous positioning results and the vehicle's prior state estimation results at the same time. The difference between the vehicle's continuous positioning results and the vehicle's position corresponding to the visual observations is determined by comparing the degree of deviation between the relative position of the vehicle reflected by the visual observations and the current positioning results to obtain the observation bias. The determination is based on the degree of conformity between the vehicle's continuous positioning results and the vehicle positioning constraints. Specifically, it is determined whether the vehicle's current position satisfies the constraint relationship formed by each virtual anchor point in the environment, and the constraint deviation is determined based on the degree of non-compliance. A time-series comparison is performed on the consistency judgment results across consecutive time points to determine the changing states of the current position deviation, observation deviation, and constraint deviation over consecutive time points. Based on these changing states, it is determined whether the vehicle's continuous positioning result at the current time point meets preset anomaly conditions. The preset anomaly conditions are as follows: At least one of the current position deviation, observation deviation, or constraint deviation shows a monotonically increasing trend over three consecutive time intervals, or the change in any deviation at the current time interval relative to the previous time interval exceeds a preset range; When the vehicle's continuous positioning results at the current moment meet the preset abnormal conditions, an observation reconstruction operation is performed to redistribute the dynamic credibility results corresponding to each input, redetermine the constraint action order of each input based on the redistributed dynamic credibility results, re-identify and filter the virtual anchor points of the environment, and update the vehicle positioning constraints. Based on the reallocated dynamic reliability results, the redefined constraint order, and the updated vehicle positioning constraints, the solution process for the vehicle position state, speed state, and heading state is re-executed to obtain and output the corrected continuous vehicle positioning results.

[0029] refer to Figure 2 A vehicle continuous positioning system based on UAV relay guidance includes: The drone relay guidance module is used to acquire drone pose information and establish a wireless communication link between the drone and the vehicle. The vehicle prior state module is used to acquire vehicle motion state information and generate prior state estimation results for the vehicle's current position. The visual observation generation module is used to generate visual motion constraint information of vehicle targets and construct visual observations based on the improved LoFTR model. The virtual anchor point construction module is used to determine the location of environmental reflections and construct vehicle positioning constraints; The state solution module is used to jointly process the vehicle's prior state estimation results, visual observations, and vehicle positioning constraints to perform iterative solution of the vehicle's position state and obtain continuous vehicle positioning results. The observation reconstruction module is used to determine the consistency of continuous vehicle positioning results and to perform observation reconstruction under abnormal conditions to maintain continuous positioning output.

[0030] Example 1: To verify the feasibility of this invention in practice, it was applied to a typical complex urban road area, including densely built-up areas, under-bridge areas, and short-distance urban tunnels, which are areas where satellite signals are easily blocked and multipath effects are significant. The test was conducted on a weekday morning during normal traffic hours, with an average vehicle speed of 30 to 45 kilometers per hour. The test vehicle was equipped with an inertial measurement unit, wheel speed sensors, and communication equipment. The UAV used a multi-rotor platform, maintaining a flight altitude of approximately 80 to 120 meters, and synchronously followed the vehicle's path.

[0031] In this scenario, a drone relay guidance node is first deployed. The drone acquires its own pose information and establishes a wireless communication link with the vehicle. During operation, the vehicle acquires inertial measurement data, wheel speed data, and steering data in real time. Based on the motion state information, it generates a priori state estimate of the vehicle's current position and transmits it to the drone via the wireless communication link. The drone continuously collects a series of images of the vehicle's location and inputs adjacent time-series images and corresponding drone pose information into an improved LoFTR model. Through the collaborative efforts of the pose guidance input unit, target region enhancement unit, and cross-temporal feature preservation unit, visual motion constraint information of the vehicle target is generated, and a visual observation of the vehicle relative to the drone is further constructed.

[0032] Wireless communication signals between the UAV and the vehicle are used to extract multipath propagation information. By analyzing the propagation characteristics at continuous moments, stable reflection paths are selected, and the corresponding environmental reflection positions are determined. These are then used as virtual environmental anchor points to construct vehicle positioning constraints. During the positioning process, a state solution structure based on dynamic credibility constraint allocation is constructed. Through a multi-source credibility evolution unit, a constraint priority scheduling unit, and a progressive state convergence unit, the vehicle's prior state estimation results, visual observations, and vehicle positioning constraints are jointly processed to achieve continuous updates of the vehicle's position, speed, and heading. When an anomaly is detected between the positioning result and the multi-source observations, an observation reconstruction operation is performed. By reallocating credibility weights and updating the virtual environmental anchor point constraints, the positioning result is restored to stability, thereby achieving continuous vehicle positioning.

[0033] To verify the beneficial effects of the present invention, the method of the present invention was compared with the traditional satellite navigation combined with inertial fusion positioning method under the same route conditions, and a high-precision reference positioning device was used as the true value for error evaluation.

[0034] Table 1 Comparison of Vehicle Positioning Performance in Complex Environments Test area Road segment length (m) Average error (m) of traditional method The average error (m) of this invention Maximum error (m) of the traditional method Maximum error (m) of this invention Number of interruptions (traditional / inventional) open road section 800 1.5 1.3 2.8 2.4 0 / 0 High-rise building area 1100 6.2 3.1 11.4 5.6 1 / 0 Under the bridge area 500 8.7 3.8 14.9 6.3 2 / 0 tunnel area 600 No continuous localization results 4.5 — 7.2 3 / 0 Comprehensive analysis of the entire road section 3000 5.2 2.9 14.9 7.2 6 / 0 As can be seen from the data in Table 1, under open road conditions, the positioning performance of the traditional method and the method of the present invention is similar. The average error of both is controlled within the range of 1 to 2 meters, and the maximum error is less than 3 meters. This indicates that under good satellite signal and sufficient observation conditions, the present invention will not reduce the original positioning performance and can maintain an accuracy level comparable to the existing technology.

[0035] In areas with high-rise buildings and under bridges, the positioning error of traditional methods increases significantly, with average errors reaching 6.2 meters and 8.7 meters respectively, and maximum errors exceeding 10 meters and even approaching 15 meters, while multiple positioning interruptions occur. In contrast, the method of this invention, under the same conditions, controls the average error to 3.1 meters and 3.8 meters respectively, with maximum errors of 5.6 meters and 6.3 meters respectively, and no positioning interruptions occur. This indicates that by introducing UAV visual observation and environmental virtual anchor point constraints, this invention effectively reduces the impact of multipath interference and signal obstruction on the positioning results, improving positioning stability in complex environments.

[0036] In tunnel areas, traditional methods cannot provide continuous positioning results due to the inability to acquire effective satellite signals. However, the method of this invention can still maintain continuous positioning, with an average error of 4.5 meters and a maximum error of 7.2 meters. This demonstrates that the invention can achieve stable positioning by relying on visual observation and virtual anchor point constraints even without satellite signals. Looking at the overall results across the entire road section, the invention reduces the overall average error from 5.2 meters to 2.9 meters, eliminating positioning interruptions. This indicates that the invention is superior to traditional methods in terms of continuity, stability, and anti-interference capabilities, effectively solving the problem of insufficient stability in continuous vehicle positioning under complex environments.

[0037] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for continuous vehicle positioning based on UAV relay guidance, characterized in that, include: Deploy drone relay guidance nodes to obtain drone pose information and establish wireless communication links between drones and vehicles; The system acquires the vehicle's motion state information, generates a priori state estimate of the vehicle's current position based on the motion state information, and sends it to the UAV relay guidance node via a wireless communication link. A continuous image sequence containing the vehicle target is obtained. Images at adjacent time points and corresponding UAV pose information are input into the improved LoFTR model to generate visual motion constraint information of the vehicle target. The visual observation of the vehicle relative to the UAV is constructed by combining the UAV pose information. Acquire multipath propagation information of wireless signals between UAV and vehicle, identify stable reflection propagation paths in continuous time and determine environmental reflection positions, and use environmental reflection positions as virtual environmental anchor points to construct vehicle positioning constraints. A state-solving module based on dynamic credibility constraint allocation is constructed to jointly process the vehicle prior state estimation results, visual observations, and vehicle positioning constraints. According to the stability of each observation source in continuous time, a corresponding credibility weight is assigned to each observation source, and the vehicle position state is iteratively solved to obtain the continuous vehicle positioning results. Consistency determination is performed on the continuous vehicle positioning results. When the consistency determination results meet the preset abnormal conditions, the observation reconstruction operation is performed. By adjusting the confidence weight or re-identifying the virtual anchor points of the environment, the continuous vehicle positioning output is maintained.

2. The method for continuous vehicle positioning based on UAV relay guidance according to claim 1, characterized in that, The acquisition of UAV pose information includes: The drone is controlled to fly to a preset airspace above the vehicle or above the vehicle's driving path, and satellite navigation positioning data, inertial measurement data, and barometric altitude data of the drone are acquired. The satellite navigation positioning data, inertial measurement data, and barometric altitude data are time-aligned, and short-term compensation is performed on the satellite navigation positioning data based on the inertial measurement data to obtain the drone's position coordinates and attitude angle information, including roll angle, pitch angle, and heading angle.

3. The method for continuous vehicle positioning based on UAV relay guidance according to claim 1, characterized in that, Establishing a wireless communication link between the drone and the vehicle includes: By scanning signals and initializing links between the UAV communication unit and the vehicle communication unit, a two-way communication connection is established between the UAV and the vehicle. Time synchronization is performed between the two communicating parties. After the communication is established, the link signal strength and communication delay information are continuously acquired. The communication link status is determined based on the signal strength and communication delay information. When the link status meets the preset conditions, the stable connection of the wireless communication link is maintained.

4. The method for continuous vehicle positioning based on UAV relay guidance according to claim 1, characterized in that, The step of generating a priori state estimate of the vehicle's current position based on motion state information and sending it to the UAV relay guidance node via a wireless communication link includes: The motion state information of the vehicle at the current moment is obtained. The motion state information includes the vehicle's inertial measurement data, wheel speed data, and steering data. The inertial measurement data, wheel speed data, and steering data are time-aligned according to a unified sampling time sequence to form the vehicle motion state sequence corresponding to the current moment. Based on the vehicle motion state sequence, extract the longitudinal motion change features, lateral motion change features, and heading change features of the vehicle in continuous time moments, and construct the vehicle short-time trajectory evolution sequence based on the longitudinal motion change features, lateral motion change features, and heading change features; The short-term trajectory evolution sequence of the vehicle is subjected to steering-sensitive discrimination processing. When the vehicle is detected to be turning, the weight of the steering data on the trajectory evolution sequence is increased. When the vehicle is detected to be traveling straight, the weight of the wheel speed data on the trajectory evolution sequence is increased, and the direction correction result corresponding to the current driving state is generated. The trajectory evolution sequence and the direction correction result are subjected to continuity constraint processing to suppress position jumps caused by instantaneous fluctuations in continuous time intervals, and the prior trajectory position at the current time is generated according to the motion continuity relationship of the vehicle in continuous time intervals. The prior trajectory position is combined with the current velocity and heading states to generate a prior state estimate of the vehicle's current position, and the prior state estimate is sent to the UAV relay guidance node.

5. A method for continuous vehicle positioning based on UAV relay guidance according to claim 1, characterized in that, The process of generating visual motion constraint information for the vehicle target and constructing a visual observation of the vehicle relative to the UAV by combining the UAV pose information includes: An improved LoFTR model is constructed, which includes a pose-guided input unit, a target region enhancement unit, a cross-temporal feature preservation unit, a visual motion constraint generation unit, and a credibility feedback correction unit. The pose guidance input unit receives images from adjacent time points, UAV pose information, and vehicle prior state estimation results. Based on the UAV pose information and vehicle prior state estimation results, it determines the predicted region of the vehicle target in the current image and generates corresponding region guidance information. The target region enhancement unit receives image features and region guidance information, performs feature enhancement on the region corresponding to the vehicle target, performs feature suppression on the non-target region, and outputs a target enhancement feature map. Position encoding is performed on the target enhanced feature map. Intra-image feature association and cross-image feature association are performed on the encoded feature map to generate cross-time-time association features. The initial matching result is determined based on the cross-time-time association features. The initial matching result is locally refined to obtain candidate matching features between the current time and the previous time. The cross-temporal feature preservation unit receives candidate matching features and matching trajectory information from continuous historical moments. It jointly determines the position changes, direction changes, and displacement continuity of each matching feature in continuous moments, retains stable matching features that meet the continuity condition, and forms a temporal stable feature set of the vehicle target. The visual motion constraint generation unit receives a time-stable feature set, extracts the displacement change trend, direction change trend and viewpoint change relationship of the vehicle target in continuous time, generates visual motion constraint information of the vehicle target, and forms a relative motion description result of the vehicle target relative to the UAV based on the visual motion constraint information. The credibility feedback correction unit receives the relative motion description results, UAV pose information, and vehicle continuous positioning results at the current moment, performs credibility correction on the visual motion constraint information, and constructs the visual observation of the vehicle relative to the UAV based on the corrected visual motion constraint information and UAV pose information. The improved LoFTR model was trained, and the visual motion constraint reconstruction error was used as the optimization target. The parameters of the pose-guided input unit, target region enhancement unit, cross-temporal feature preservation unit, visual motion constraint generation unit, and credibility feedback correction unit were continuously optimized.

6. A vehicle continuous positioning system and method based on UAV relay guidance according to claim 1, characterized in that, The method of constructing vehicle positioning constraints by using environmental reflection locations as virtual environmental anchor points includes: The system acquires multipath information of wireless propagation signals between the UAV and the vehicle at continuous time intervals, extracts the propagation delay change information and signal strength change information of each propagation path at continuous time intervals, and forms a multipath time sequence information according to the corresponding relationship of the changes of the propagation path at continuous time intervals. Based on the multipath timing information sequence, propagation paths with continuous propagation delay changes and stable signal strength changes in consecutive time moments are screened to determine stable reflection propagation paths, and the corresponding environmental reflection positions are determined based on the propagation characteristics of stable reflection propagation paths in consecutive time moments. By jointly associating the environmental reflection positions with the current pose information of the UAV and the prior state estimation results of the vehicle, the spatial distribution relationship of each environmental reflection position relative to the vehicle's driving direction is determined, and environmental reflection positions located in the current forward region or lateral continuous region of the vehicle's current driving direction are screened to form a candidate set of virtual environmental anchor points. The environmental reflection positions in the candidate set of virtual environmental anchor points are continuously calibrated. Based on the degree of constraint offset of the vehicle prior state estimation result at the same environmental reflection position in consecutive time moments, environmental reflection positions with continuous constraint offset and consistent direction are retained as valid virtual environmental anchor points. Establish corresponding constraints between the effective virtual anchor points of the environment and the prior state estimation results of the vehicle's current position, so that each effective virtual anchor point of the environment forms an independent position convergence constraint on the vehicle's current position. The order of constraint action is determined according to the stability of each effective virtual anchor point in continuous time, and vehicle positioning constraints for continuous vehicle positioning are generated.

7. A method for continuous vehicle positioning based on UAV relay guidance according to claim 1, characterized in that, Based on the stability of each observation source over consecutive time intervals, a corresponding confidence weight is assigned to each observation source, and the vehicle position state is iteratively solved to obtain the continuous vehicle positioning results, including: A state-solving module based on dynamic credibility constraint allocation is constructed. The state-solving module includes a multi-source credibility evolution unit, a constraint priority scheduling unit, and a progressive state convergence unit. The multi-source credibility evolution unit receives the vehicle's prior state estimation results, visual observations, and vehicle positioning constraints, and forms corresponding prior state inputs, visual constraint inputs, and virtual anchor point constraint inputs, respectively. The multi-source credibility evolution unit jointly determines the continuity, directional consistency, and constraint offset stability of the changes in prior state input, visual constraint input, and virtual anchor point constraint input in continuous time intervals, generates dynamic credibility results corresponding to each input, and forms a credibility evolution sequence for each input according to the smoothness of credibility changes in continuous time intervals. The constraint priority scheduling unit receives the confidence evolution sequence and sorts the dynamic confidence results of the prior state input, visual constraint input, and virtual anchor point constraint input in descending order to determine the constraint action order of each input. Then, the corresponding inputs are selected in sequence according to the constraint action order to participate in the solution at the current position, forming the hierarchical constraint combination result at the current time. The progressive state convergence unit receives the results of the hierarchical constraint combination. It first uses the input with the highest priority to generate the initial solution result of the vehicle's current position, and then introduces the remaining inputs in sequence to correct the initial solution result layer by layer. This allows the vehicle's position state, velocity state, and heading state to gradually converge during the continuous correction process, thus obtaining the vehicle state result at the current moment. The vehicle state results are fed back to the multi-source credibility evolution unit to update the dynamic credibility results and constraint action order of each input at the next moment. The vehicle position state, velocity state and heading state after convergence at the current moment are output as the vehicle continuous positioning results.

8. A method for continuous vehicle positioning based on UAV relay guidance according to claim 1, characterized in that, The observation reconstruction operation, which maintains continuous vehicle positioning output by adjusting confidence weights or re-identifying virtual environmental anchor points, includes: Based on the vehicle's continuous positioning results, the vehicle's prior state estimation results, visual observations, and vehicle positioning constraints, the current position deviation, observation deviation, and constraint deviation are determined to form the consistency judgment result at the current moment. The consistency judgment results in consecutive time moments are compared in time sequence to determine the change state of the current position deviation, observation deviation and constraint deviation in consecutive time moments, and the vehicle continuous positioning result at the current time moment is judged to meet the preset abnormal conditions based on the change state. When the vehicle's continuous positioning results at the current moment meet the preset abnormal conditions, an observation reconstruction operation is performed to redistribute the dynamic credibility results corresponding to each input, redetermine the constraint action order of each input based on the redistributed dynamic credibility results, re-identify and filter the virtual anchor points of the environment, and update the vehicle positioning constraints. Based on the reallocated dynamic reliability results, the redefined constraint order, and the updated vehicle positioning constraints, the solution process for the vehicle position state, speed state, and heading state is re-executed to obtain and output the corrected continuous vehicle positioning results.

9. A vehicle continuous positioning system based on UAV relay guidance, wherein the vehicle continuous positioning method based on UAV relay guidance as described in any one of claims 1 to 8 is characterized in that, include: The drone relay guidance module is used to acquire drone pose information and establish a wireless communication link between the drone and the vehicle. The vehicle prior state module is used to acquire vehicle motion state information and generate prior state estimation results for the vehicle's current position. The visual observation generation module is used to generate visual motion constraint information of vehicle targets and construct visual observations based on the improved LoFTR model. The virtual anchor point construction module is used to determine the location of environmental reflections and construct vehicle positioning constraints; The state solution module is used to jointly process the vehicle's prior state estimation results, visual observations, and vehicle positioning constraints to perform iterative solution of the vehicle's position state and obtain continuous vehicle positioning results. The observation reconstruction module is used to determine the consistency of continuous vehicle positioning results and to perform observation reconstruction under abnormal conditions to maintain continuous positioning output.

Images