Unmanned bus high-precision positioning method based on SLAM map

Abstract

The invention discloses an unmanned bus high-precision positioning method based on an SLAM map, and relates to the technical field of unmanned driving, and the method comprises the following steps: obtaining laser radar data, inertial measurement data, odometer data and visual data of an unmanned bus; analyzing a vehicle control instruction, simulating the movement track of the unmanned bus through a particle filter algorithm, and generating a pose hypothesis set; performing position matching on the laser radar data and a pre-constructed SLAM map, and performing auxiliary positioning in a low-visibility scene to obtain a position matching result; initial static map construction and dynamic obstacle detection are carried out, and a dynamic and static fusion SLAM map is generated; and state monitoring is carried out, and a multi-stage fault recovery strategy is triggered. According to the method, through dynamic fusion of timestamp alignment and Kalman filtering, severe environment parameter adjustment and a multi-stage fault recovery mechanism, the positioning accuracy and the system stability of the driverless bus are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of driverless technology, and more specifically, to a high-precision positioning method for driverless buses based on SLAM maps. Background Art

[0002] As an important part of intelligent transportation systems, driverless technology is gradually changing the face of urban public transportation. Especially in the field of driverless buses, by integrating advanced sensors, computing platforms, and control algorithms, it aims to achieve safe and efficient urban public transportation services. It needs to accurately obtain its own position in various complex environments for precise path planning and obstacle avoidance operations.

[0003] In the prior art, multi-source data fusion is a key link in achieving precise positioning of driverless buses. However, traditional methods often rely on simple superposition or asynchronous processing of sensor data, which leads to data accumulation error problems caused by time asynchronization, affecting the stability and real-time performance of the positioning system. Moreover, existing fusion algorithms fail to fully consider the noise differences between sensor data, resulting in inaccurate motion state estimation and limiting the safety and reliability of driverless buses.

[0004] Secondly, for special situations such as low visibility environments like haze and rainy days, existing positioning technologies mainly rely on a single lidar for environmental perception. However, in adverse weather conditions, the effective detection range and accuracy of lidar will be greatly affected. It also lacks effective auxiliary means to make up for the positioning inaccuracy problem caused by the performance degradation of lidar, making it difficult for driverless buses to maintain normal operating capabilities under such conditions.

[0005] In addition, for the recognition and handling of dynamic obstacles, most current technologies adopt static map construction and simple obstacle classification methods, which are difficult to adapt to dynamic elements such as pedestrians and vehicles that frequently appear on urban roads. This kind of limitation not only reduces the efficiency of path planning but may also lead to traffic accidents due to misjudgment. At the same time, existing systems lack a flexible recovery mechanism when encountering failures, and an error may cause the entire system to fail.

[0006] Therefore, how to design a high-precision positioning method for driverless buses based on SLAM maps, which can effectively fuse multi-source data, improve the positioning accuracy in low visibility environments, and have the ability to intelligently process dynamic obstacles and quickly recover from failures, is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0007] In view of this, the present invention provides a high-precision positioning method for driverless buses based on SLAM maps, which provides more reliable and efficient positioning and navigation support for driverless buses through precise time synchronization and data fusion, an adaptive low visibility enhancement strategy, and a dynamic obstacle hierarchical modeling and multi-level fault tolerance mechanism, so as to meet the safety operation requirements in complex urban environments.

[0008] To achieve the above object, the present invention adopts the following technical solutions:

[0009] A high-precision positioning method for driverless buses based on SLAM maps, comprising the following steps:

[0010] S1. Obtain lidar data, inertial measurement data, odometer data, and visual data of the driverless bus, perform timestamp synchronization and integrity verification, and fuse the inertial measurement data and the odometer data;

[0011] S2. Parse the vehicle control commands using the fused data, simulate the motion trajectory of the driverless bus through the particle filter algorithm, and generate a pose hypothesis set;

[0012] S3. Based on the pose hypothesis set, perform position matching on the lidar data and the pre-constructed SLAM map, and perform auxiliary positioning in low visibility scenarios to obtain a position matching result;

[0013] S4. Through the position matching result, perform initial static map construction and dynamic obstacle detection to generate a dynamic and static fusion SLAM map;

[0014] S5. Perform state monitoring under the dynamic and static fusion SLAM map and trigger a multi-level fault recovery strategy.

[0015] Further, in the step S1, the timestamp synchronization and integrity verification include:

[0016] Timestamp synchronization: Taking the timestamp of the lidar data as a reference, align the inertial measurement data and the odometer data to the same time series through the linear interpolation algorithm;

[0017] Integrity verification: Verify that the number of points in each frame of lidar data > 500, the absolute value of the acceleration of the inertial measurement data ≤ 30 ± m / s 2 , the absolute value of the odometer data ≤ 15 m / s; if the data is incomplete, the missing data is supplemented by combining the time interpolation algorithm.

[0018] Further, in the step S1, the Kalman filter algorithm is used to fuse the inertial measurement data and the odometer data;

[0019] The state prediction equation of the Kalman filter algorithm is expressed as:

[0020] x' k = A·x' k−1 + B·u k ;

[0021] wherein, x' k represents the state prediction value at time k, and x' k−1 represents the state prediction value at time k−1, u k represents the control input vector, A represents the state transition matrix, and B represents the control input matrix;

[0022] The state update equation of the Kalman filter algorithm is expressed as:

[0023] x k = x' k + K·(z k - H·x' k );

[0024] wherein, x k represents the state update value at time k, z k represents the state measurement value at time k, K represents the Kalman gain weight matrix, and H represents the observation matrix.

[0025] Furthermore, in S2, parsing the vehicle control command includes:

[0026] According to the speed v t and the steering angular velocity ω t , calculating the displacement increment Δx and the angle increment Δθ; and setting the maximum vehicle speed to 15 m / s and the maximum steering angular velocity to ±0.5 rad / s. If the speed or the steering angular velocity exceeds the threshold, a limit constraint is triggered.

[0027] Furthermore, in S2, the calculation equation for the movement of each particle in the particle filter algorithm is:

[0028] x i,t = x i,t-1 + Δx + N 1 (0, σ x );

[0029] θ i,t = θ i,t-1 + Δθ + N 2 (0, σ θ );

[0030] wherein, x i,t represents the position coordinate of particle i at time t, x i,t-1 represents the position coordinate of particle i at time t-1, N 1 represents a normal distribution with a mean of 0 and a standard deviation of σx Position noise; θ i,t Represents the angular coordinate of particle i at time t, θ i,t-1 Represents the angular coordinate of particle i at time t-1, N 2 Represents angular noise with a mean of 0 and a standard deviation of σ θ of angular noise.

[0031] Furthermore, in the step S3, performing position matching between the lidar data and the pre-constructed SLAM map includes: mapping the lidar data to a grid map using the NDT algorithm and calculating the position matching probability of each particle :

[0032] ;

[0033] where z t represents the current observed point cloud, h(x i,t ) represents the predicted point cloud corresponding to the position coordinates of particle i at time t, ||·|| represents the Euclidean distance, exp(·) represents the exponential function, and σ represents the standard deviation of the matching error.

[0034] Furthermore, in the step S3, performing auxiliary positioning in a low visibility scenario includes:

[0035] Limiting the effective point cloud detection distance to 20 m and adjusting the lidar acquisition frequency to 15 Hz or 20 Hz in combination with the weather conditions;

[0036] Combining visual data to extract road markings and traffic signs, generating auxiliary positioning feature points, and performing weighted fusion with the position matching results.

[0037] Furthermore, in the step S4, constructing the initial static map includes:

[0038] Projecting the initial point cloud onto a two-dimensional grid map with a resolution of 0.1 m, marking obstacles as non-passable areas; if the map construction fails, reducing the resolution to 0.2 m and increasing the number of point cloud acquisitions; if it still fails, prohibiting the vehicle from starting and triggering an alarm.

[0039] Furthermore, in the step S4, detecting dynamic obstacles includes:

[0040] Extracting dynamic obstacles through point cloud difference and setting a speed threshold for classification;

[0041] Using the YOLO algorithm to identify the types of obstacles in the visual data and dynamically assigning weight priorities according to the distance;

[0042] And, increasing the lower limit of the dynamic obstacle speed threshold to 0.5 m / s in rainy weather scenarios and correcting the vehicle slip error through the inertial measurement unit.

[0043] Further, in S5, the multi-level fault recovery strategy includes:

[0044] If the mean particle weight < 0.01, the sensor data loss lasts for ≥ 1 second, or the map update delay ≥ 500 ms, it is determined that the positioning fails; then reset the particle distribution to the historical highest weight state, and remove the latest dynamic obstacle marker and update to the previous map version;

[0045] If the map update fails, enable the block storage mechanism and only update the local map within a range of 30 meters around.

[0046] Through the above technical solutions, compared with the prior art, the technical solutions of the present invention have the following beneficial effects:

[0047] 1. This method aligns multi-sensor data through a linear interpolation algorithm based on the lidar timestamp, and combines the Kalman filter algorithm to fuse the inertial measurement unit and odometer data, significantly improving the data consistency and the stability of motion state estimation. Compared with the asynchronous or simple superposition fusion method of sensor data in the prior art, this method effectively solves the cumulative error problem caused by the time asynchrony of multi-source data, and enhances the real-time performance and reliability of the positioning system.

[0048] 2. For harsh environments such as haze and rainy days, it proposes to limit the effective detection distance of the lidar, dynamically adjust the point cloud acquisition frequency, and fuse visual data to extract road markings and traffic signs as auxiliary positioning feature points. Through multi-modal data complementarity and parameter adaptive adjustment, it improves the environmental adaptability in low visibility scenarios.

[0049] 3. Combining point cloud difference, speed threshold classification and YOLO object detection algorithm, it realizes the accurate identification of dynamic obstacles and dynamic weight assignment of priorities; at the same time, it designs a multi-level fault recovery strategy, including hierarchical response measures such as particle weight reset, local map rollback, and emergency stop. Through the hierarchical modeling of dynamic obstacles and multi-level fault tolerance logic, it greatly improves the system safety and anti-interference ability in complex urban road scenarios. Description of the Drawings

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained according to the provided drawings.

[0051] Figure 1Flowchart of a high-precision positioning method for driverless buses based on SLAM maps provided by embodiments of the present invention;

[0052] Figure 2 Flowchart of a multi-level fault recovery strategy provided by embodiments of the present invention. Detailed implementation manners

[0053] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0054] As Figure 1 shown, this embodiment provides a high-precision positioning method for driverless buses based on SLAM maps, including:

[0055] S1. Obtain the lidar data, inertial measurement data, odometer data, and visual data of the driverless bus, perform timestamp synchronization and integrity verification, and fuse the inertial measurement data and odometer data;

[0056] S2. Use the fused data to parse vehicle control commands, simulate the motion trajectory of the driverless bus through the particle filter algorithm, and generate a pose hypothesis set;

[0057] S3. Based on the pose hypothesis set, perform position matching on the lidar data and the pre-constructed SLAM map, and perform auxiliary positioning in low visibility scenarios to obtain a position matching result;

[0058] S4. Through the position matching result, perform initial static map construction and dynamic obstacle detection to generate a dynamic and static fusion SLAM map;

[0059] S5. Perform state monitoring under the dynamic and static fusion SLAM map and trigger a multi-level fault recovery strategy.

[0060] This method realizes multi-source data fusion through linear interpolation of the lidar timestamp benchmark combined with Kalman filtering, improving the stability and real-time performance of motion state estimation. For low visibility scenarios, it adaptively adjusts lidar detection parameters and fuses visual data to enhance positioning robustness. In the processing of dynamic obstacles, it uses point cloud difference, speed threshold classification, and YOLO detection for accurate identification, and designs a multi-level fault recovery strategy to improve system safety and anti-interference ability.

[0061] The following further elaborates on each step in the above method:

[0062] In this embodiment S1, the lidar data, inertial measurement data, odometer data, and visual data of the driverless bus are acquired, timestamp synchronization and integrity verification are performed, and the inertial measurement data and odometer data are fused.

[0063] Specifically, the specific acquisition methods of the above data are as follows:

[0064] Lidar data: The surrounding environment point cloud data is collected in real time by the lidar. Each frame of point cloud data contains three-dimensional coordinates and reflection intensity information. The sampling frequency is 10 times per second, the point cloud resolution is 0.1 meter, and the data format is pcd.

[0065] Inertial measurement data: The information of the vehicle's acceleration and angular velocity is collected by the inertial measurement unit IMU. The sampling frequency is 100 Hz, the data format is 6 dimensions, with a timestamp attached, and the format is json.

[0066] Odometer data: The vehicle's driving speed (error accuracy is ±0.05) and steering angular velocity (±0.01) are collected by the odometer, and the collection frequency is 50 Hz.

[0067] Visual data: A high-definition camera is used to provide supplementary positioning data for the autonomous driving bus in complex scenarios such as haze environment, urban traffic main roads, and complex intersections.

[0068] When the lidar data is lost, the driverless bus will automatically switch to using inertial measurement data and odometer data for compensation to maintain the basic navigation function of the vehicle. In addition, if the acceleration detected by the inertial measurement unit IMU exceeds the set threshold, or the speed change shown by the odometer is abnormal, the vehicle deceleration or suspension operation will be triggered. This mechanism ensures that the vehicle can drive safely even when part of the sensor data is lost and effectively prevents risks caused by abnormal data.

[0069] Furthermore, the timestamp synchronization and integrity verification include:

[0070] Timestamp synchronization: Based on the timestamp of the lidar data, the inertial measurement data and odometer data are aligned to the same time series through the linear interpolation algorithm.

[0071] Integrity verification: Check the timestamp of each frame of data. The time interval of each frame of lidar point cloud ≤ 100 ms, the time interval of each frame of inertial measurement unit data ≤ 10 ms, and the time interval of each frame of odometer data ≤ 20 ms; and verify that the number of points in each frame of lidar data > 500, the absolute value of the acceleration of inertial measurement data ≤ 30 ± m / s 2 , the absolute value of odometer data ≤ 15 m / s; if the data is incomplete, the missing data is complemented by combining the time interpolation algorithm.

[0072] Furthermore, the Kalman filter algorithm is used to fuse the inertial measurement data and the odometer data;

[0073] The state prediction equation of the Kalman filter algorithm is expressed as:

[0074] x’ k =A·x’ k−1 +B·u k ;

[0075] where, x’ k represents the state prediction value at time k, x’ k−1 represents the state prediction value at time k−1, u k represents the control input vector, A represents the state transition matrix, and B represents the control input matrix;

[0076] The state update equation of the Kalman filter algorithm is expressed as:

[0077] x k =x’ k +K·(z k -H·x’ k );

[0078] where, x k represents the state update value at time k, z k represents the state measurement value at time k, K represents the Kalman gain weight matrix, and H represents the observation matrix.

[0079] Furthermore, if the fusion fails, the lidar point cloud is directly used to generate a temporary position. In addition, if the Kalman gain weight matrix K is unstable (i.e., the numerical value diverges) during the Kalman filter update process, the filter state is re-initialized.

[0080] This step solves the multi-sensor timing difference problem through the timestamp synchronization and data compensation mechanism; uses the Kalman filter to improve the accuracy and real-time performance of the motion state estimation; automatically switches to the backup data source when the sensor fails, and combines the dynamic threshold to trigger deceleration or pause operations, significantly enhancing the robustness, fault tolerance, and driving safety of the system.

[0081] In this embodiment S2, the fused data is used to parse the vehicle control commands, and the particle filter algorithm is used to simulate the motion trajectory of the driverless bus to generate a pose hypothesis set;

[0082] Among them, parsing the vehicle control commands includes:

[0083] According to the speed v t and the steering angular velocity ω t, calculate the displacement increment Δx and the angular increment Δθ; and set the maximum vehicle speed to 15 m / s and the maximum steering angular velocity to ±0.5 rad / s. If the speed or the steering angular velocity exceeds the threshold, a limit constraint is triggered.

[0084] Further, in the S2, the calculation equations for the movement of each particle in the particle filter algorithm are:

[0085] x i,t =x i,t-1 +Δx+N 1 (0,σ x );

[0086] θ i,t =θ i,t-1 +Δθ+N 2 (0,σ θ );

[0087] Among them, x i,t represents the position coordinate of particle i at time t, x i,t-1 represents the position coordinate of particle i at time t - 1, N 1 represents the position noise with a mean of 0 and a standard deviation of σ x ; θ i,t represents the angular coordinate of particle i at time t, θ i,t-1 represents the angular coordinate of particle i at time t - 1, N 2 represents the angular noise with a mean of 0 and a standard deviation of σ θ .

[0088] Further, the standard deviation σ x of the position noise is set to 0.05 meters, and the standard deviation σ θ of the angular noise is set to 0.01 radians.

[0089] In this step, based on the fused data, vehicle control commands are parsed, the motion trajectory is simulated through the particle filter algorithm, a position noise and an angular noise model are introduced to enhance the prediction diversity, and a limit constraint is combined to ensure driving safety. Among them, the particle filter algorithm effectively processes the nonlinear motion model, and the optimization of the noise parameters improves the coverage and robustness of the hypothesis set; the dynamic constraint mechanism suppresses the overspeed risk in real time, ensuring the accuracy of the trajectory prediction and the stability of vehicle control, and providing a reliable pose hypothesis basis for subsequent positioning and decision-making.

[0090] In the S3 of this embodiment, based on the pose hypothesis set, the lidar data is position-matched with the pre-constructed SLAM map, and auxiliary positioning is performed in a low visibility scenario to obtain a position matching result;

[0091] Among them, the position matching of the lidar data with the pre-constructed SLAM map includes:

[0092] In a normal environment, points within 30 - 50 meters in front of the vehicle are retained, and the NDT algorithm is used to map the lidar data to a grid map, and the position matching probability of each particle is calculated. :

[0093] ;

[0094] Among them, z t represents the current observed point cloud, h(x i,t ) represents the predicted point cloud corresponding to the position coordinates of particle i at time t, ||·|| represents the Euclidean distance, exp(·) represents the exponential function, and σ represents the standard deviation of the matching error, which is specifically set to 0.1m.

[0095] Furthermore, for auxiliary positioning in low visibility scenarios, it includes:

[0096] The effective point cloud detection distance is limited to 20 m, and the lidar acquisition frequency is adjusted according to the weather conditions to 15 Hz (haze scenario) or 20 Hz (rainy scenario);

[0097] Combined with visual data, road markings and traffic signs are extracted to generate auxiliary positioning feature points, which are weighted and fused with the position matching results.

[0098] In addition, for the complex scenario of an autonomous driving bus running in rainy days, since the raindrops have noise interference on the lidar reflection signal, making the camera vision blurred, and the road surface water may cause the vehicle to slip, which has a certain impact on the accuracy of the vehicle motion model. The correction effect of the inertial measurement unit on the vehicle attitude is enhanced to compensate for the data error of the odometer. And further update and normalize the particle weights to reflect the environmental characteristics.

[0099] This step efficiently matches the lidar data with the pre - built SLAM map through the NDT algorithm, and adaptively adjusts according to the weather, and optimizes the positioning result by fusing visual feature points. It integrates multi - source data through a weighted fusion strategy to effectively handle the problems of rainy - day radar noise, camera blur, and vehicle slip; while the dynamic particle weight update and IMU attitude correction compensate for the odometer error, significantly improving the positioning accuracy and environmental adaptability in low visibility scenarios, and ensuring the stable operation of the autonomous driving bus in complex weather.

[0100] In this embodiment S4, through the position matching result, an initial static map is constructed and dynamic obstacles are detected to generate a dynamic - static fusion SLAM map;

[0101] Among them, the construction of the initial static map includes:

[0102] Project the initial point cloud onto a 2D grid map with a resolution of 0.1 m, and mark obstacles as non-traversable areas; if the map construction fails, reduce the resolution to 0.2 m and increase the number of point cloud acquisitions; if it still fails, prohibit the vehicle from starting and trigger an alarm.

[0103] Further, in S4, the dynamic obstacle detection includes:

[0104] Compare the current frame point cloud with the previous frame point cloud, extract dynamic obstacles through point cloud difference, and set a speed threshold for classification; specifically, for classifying obstacles, when an autonomous driving bus is running on a conventional road, judge dynamic objects through the speed threshold, and set the recognition speed range for low-speed targets (pedestrians) and high-speed targets (vehicles). Remove the dynamic obstacle area from the static map and mark the position of the dynamic obstacle in the local temporary map to avoid the planned path passing through this area.

[0105] Further, in the autonomous driving bus operation scenario in obstacle-intensive areas such as urban main roads, based on lidar combined with high-resolution video for recognition, use the YOLO algorithm to identify and distinguish dynamic obstacles such as pedestrians, motor vehicles, electric bicycles, and bicycles, and adjust the priority of the dynamic obstacles, giving higher weights to the targets close to the autonomous driving bus; the priority weighting formula is:

[0106] ;

[0107] where, w priority represents the priority weight of the dynamic obstacle, d represents the Euclidean distance between the dynamic obstacle and the autonomous driving bus, ϵ represents a small constant to prevent the weight from becoming infinite when the obstacle is infinitely close to the vehicle, and ϵ is set to 0.1 m;

[0108] Also, increase the lower limit of the dynamic obstacle speed threshold to 0.5 m / s in rainy weather scenarios, and correct the vehicle slip error through the inertial measurement unit.

[0109] In addition, if the obstacle detection is misjudged (such as a dynamic object being misidentified as static), introduce lidar time series difference for secondary detection. If it still fails, by default, mark the suspicious area as a dangerous area and do not allow passage.

[0110] Further, perform real-time map incremental update to ensure the vehicle's timely response to environmental changes; during the process of adding newly detected dynamic obstacle information to the map, set weights for the newly added areas. The rules for setting the weights of newly added obstacles are:

[0111] ;

[0112] This step accurately separates dynamic and static obstacles through point cloud difference and speed threshold classification, combines the YOLO algorithm to realize the recognition of multi-category dynamic targets such as pedestrians and vehicles, and optimizes the path planning based on the distance priority weight. It fuses lidar and visual data in dynamic obstacle detection to improve classification accuracy and real-time performance; while the static map adaptive resolution adjustment and incremental update mechanism ensure the reliability of the map; the speed threshold optimization and secondary detection mechanism in rainy weather scenarios reduce the risk of misjudgment, combined with the dangerous area marking and weight rules, significantly enhance the dynamic response ability and driving safety in complex road environments.

[0113] In this embodiment S5, the state monitoring under the dynamic and static fusion SLAM map is carried out, and a multi-level fault recovery strategy is triggered.

[0114] As Figure 2 shown, the multi-level fault recovery strategy includes:

[0115] If the mean particle weight < 0.01, the sensor data loss lasts for ≥ 1 second, or the map update delay ≥ 500 ms, it is determined that the positioning fails; then reset the particle distribution to the historical highest weight state, and remove the latest dynamic obstacle marker and update it to the previous map version;

[0116] If the map update fails, enable the block storage mechanism and only update the local map within 30 meters around.

[0117] Furthermore, the multi-level fault recovery strategy also includes:

[0118] If a speed anomaly obstacle is detected, trigger a voice alarm, emergency stop, and remote notification to the monitoring center.

[0119] This step triggers a multi-level recovery strategy by real-time monitoring of particle weight, sensor status, and map update delay, and starts emergency stop and remote warning for speed anomaly obstacles. Its dynamic rollback mechanism quickly restores the positioning function combined with historical states, and the block update reduces the computational load; the multi-level fault determination and response significantly improve the system fault tolerance and real-time performance, ensuring the continuous and stable operation and active safety protection of the autonomous driving bus in complex scenarios.

[0120] This embodiment proposes a high-precision positioning method for driverless buses based on SLAM maps. Through multi-source data fusion and timestamp synchronization, a pose hypothesis set is generated by combining particle filters, and the NDT algorithm is used to achieve efficient matching between lidar and maps. For low visibility scenarios, radar parameters are adaptively adjusted and visual data is fused to enhance positioning robustness; dynamic obstacles are accurately detected through point cloud differencing, the YOLO algorithm, and priority weights, and a dynamic and static fusion map is constructed. A multi-level fault recovery strategy monitors the system status in real time and quickly responds to positioning failures or abnormal obstacles. This method comprehensively improves the positioning accuracy, dynamic obstacle handling ability, and system fault tolerance in complex environments, ensuring the safe and stable operation of autonomous driving buses in adverse weather and dense scenarios.

[0121] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and reference can be made to the description in the method part for related parts.

[0122] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A high-precision positioning method for driverless buses based on SLAM maps, characterized in that: The following steps are involved: S1. Obtain the lidar data, inertial measurement data, odometer data and visual data of the driverless bus, perform timestamp synchronization and integrity verification, and fuse the inertial measurement data with the odometer data; S2, using the fused data to analyze vehicle control instructions, simulating the motion trajectory of the driverless bus through the particle filter algorithm, and generating a pose hypothesis set; S3. Based on the pose hypothesis set, position matching is performed on the laser radar data and the pre-built SLAM map, and auxiliary positioning is performed in a low-visibility scene to obtain a position matching result; S4. Based on the position matching results, an initial static map is constructed and dynamic obstacle detection is performed to generate a dynamic-static fusion SLAM map; S5: Perform status monitoring under the dynamic and static fusion SLAM map and trigger a multi-level fault recovery strategy.

2. The high-precision positioning method for an unmanned bus based on a SLAM map according to claim 1 is characterized in that: In S1, performing timestamp synchronization and integrity verification includes: Timestamp synchronization: Based on the timestamp of the lidar data, the inertial measurement data and odometer data are aligned to the same time series through a linear interpolation algorithm; Integrity check: verify that the number of laser radar data points per frame is >500, and the absolute value of the acceleration of the inertial measurement data is ≤30±m / s 2 , the absolute value of the odometer data is ≤15m / s; if the data is incomplete, the time interpolation algorithm is used to complete the missing data.

3. The high-precision positioning method for an unmanned bus based on a SLAM map according to claim 1 is characterized in that: In S1, the Kalman filter algorithm is used to fuse the inertial measurement data and the odometer data; The state prediction equation of the Kalman filter algorithm is expressed as: x’ k =A·x’ k−1 +B·u k ; Among them, x' k represents the state prediction value at time k, x' k−1 represents the state prediction value at time k−1, u k represents the control input vector, A represents the state transfer matrix, and B represents the control input matrix; The state update equation of the Kalman filter algorithm is expressed as: x k =x’ k +K·(z k -H·x’ k ); Among them, x k represents the state update value at time k, z k represents the state measurement value at time k, K represents the Kalman gain weight matrix, and H represents the observation matrix.

4. The high-precision positioning method for an unmanned bus based on a SLAM map according to claim 1, characterized in that: In S2, vehicle control command parsing is performed, including: According to the speed v t and the steering angular velocity ω t , calculate the displacement increment Δx and the angle increment Δθ; and set the maximum vehicle speed to 15 m / s and the maximum steering angular velocity to ±0.5 rad / s. If the speed or steering angular velocity exceeds the threshold, the limit constraint is triggered.

5. The high-precision positioning method for an unmanned bus based on a SLAM map according to claim 1 is characterized in that: In S2, the calculation equation for the motion of each particle in the particle filter algorithm is: x i,t =x i,t-1 +Δx+N1(0,σ x ); i i,t =θ i,t-1 +Δθ+N2(0,σ θ ); Among them, x i,t represents the position coordinates of particle i at time t, x i,t-1 represents the position coordinates of particle i at time t-1, N1 represents the mean value is 0 and the standard deviation is σ x The position noise of i,t represents the angular coordinate of particle i at time t, θ i,t-1 represents the angular coordinate of particle i at time t-1, N2 represents the mean value of 0 and the standard deviation of σ θ angular noise.

6. The high-precision positioning method for an unmanned bus based on a SLAM map according to claim 1, characterized in that: In S3, the laser radar data is matched with the pre-built SLAM map, including: using the NDT algorithm to map the laser radar data to the grid map, and calculating the position matching probability of each particle. : ； Among them, z t Represents the current observation point cloud, h(x i,t ) represents the predicted point cloud corresponding to the position coordinates of particle i at time t, ||·|| represents the Euclidean distance, exp(·) represents the exponential function, and σ represents the standard deviation of the matching error.

7. The high-precision positioning method for an unmanned bus based on a SLAM map according to claim 1, characterized in that: In S3, auxiliary positioning in a low-visibility scenario is performed, including: Limit the effective point cloud detection distance to 20 m, and adjust the lidar acquisition frequency to 15 Hz or 20 Hz based on weather conditions; Road markings and traffic signs are extracted by combining visual data, and auxiliary positioning feature points are generated, which are weighted and fused with the position matching results.

8. The high-precision positioning method for an unmanned bus based on a SLAM map according to claim 1, characterized in that: In S4, the initial static map construction includes: The initial point cloud is projected onto a two-dimensional grid map with a resolution of 0.1 m, and obstacles are marked as inaccessible areas. If the map construction fails, the resolution is reduced to 0.2 m and the number of point cloud collections is increased. If it still fails, the vehicle is prohibited from starting and an alarm is triggered.

9. The high-precision positioning method for an unmanned bus based on a SLAM map according to claim 1, characterized in that: In S4, dynamic obstacle detection includes: Extract dynamic obstacles through point cloud difference and set speed threshold for classification; The YOLO algorithm is used to identify obstacle types in visual data and dynamically assign priorities based on distance; In addition, the lower limit of the dynamic obstacle speed threshold in rainy scenarios is increased to 0.5m / s, and the vehicle slip error is corrected through the inertial measurement unit.

10. The high-precision positioning method for an unmanned bus based on a SLAM map according to claim 1, characterized in that: In S5, the multi-level fault recovery strategy includes: If the particle weight mean is <0.01, sensor data loss lasts ≥1 second, or map update delay is ≥500 ms, it is considered as positioning failure. The particle distribution is reset to the highest weight state in history, and the latest dynamic obstacle mark is removed and updated to the previous map version. If the map update fails, the block storage mechanism is enabled to update only the local map within a 30-meter radius.

Images