Environment perception method based on laser radar and high-precision map

Abstract

The invention relates to an environment sensing method based on a laser radar and a high-precision map, and relates to the technical field of automatic driving. The method comprises the following steps: acquiring environmental point cloud data in real time through a laser radar; dividing a region of interest based on a high-precision map; performing target identification and classification on the environment point cloud data in the region of interest; dividing the targets in the region of interest to a dynamic barrier layer or a static barrier layer according to the motion attributes, geometric characteristics and point cloud density of the targets; and based on the road attribute information of the high-precision map, target classification and tracking results are optimized, and path planning and decision making are assisted. According to the invention, the environment perception capability in the vehicle driving process can be improved, the calculation burden is effectively reduced, the false detection rate is reduced, and the adaptability and decision reliability of the system in a complex scene are improved.

Description

Technical Field

[0001] The present application relates to the field of autonomous driving technology, and in particular to an environment perception method based on laser radar and high-precision maps. Background Art

[0002] At present, in autonomous driving systems, environmental perception technology is one of the core modules for achieving safe and reliable operation. Existing environmental perception methods usually rely on lidar and image sensors to provide dynamic and static target information through target detection and classification. However, limited by the way of processing environmental point cloud data, existing technologies often have problems with low classification accuracy and high computational complexity in complex scenarios. For example, when faced with a mixed distribution of dynamic and static targets or irregular objects, existing methods find it difficult to effectively distinguish target categories, resulting in deviations in path planning and obstacle avoidance decisions. In addition, redundant data processing increases the computational burden of the system, further affecting real-time performance. Therefore, there is an urgent need for an improved method that can improve the accuracy of environmental perception and optimize computational efficiency to meet the complex needs of autonomous driving. Summary of the invention

[0003] Based on this, it is necessary to provide an environment perception method based on laser radar and high-precision map to address the above technical problems. The method includes:

[0004] Collect environmental point cloud data in real time through LiDAR;

[0005] Divide the area of ​​interest based on high-precision maps;

[0006] Performing target recognition and classification on the environmental point cloud data within the region of interest;

[0007] According to the motion attributes, geometric characteristics and point cloud density of the target, the target in the region of interest is divided into a dynamic obstacle layer or a static obstacle layer;

[0008] The classification and tracking results of the target are optimized based on the road attribute information of the high-precision map, and path planning and decision-making are assisted.

[0009] As an optional implementation, the dividing of the region of interest based on the high-precision map includes:

[0010] Reading global road boundary information in the high-precision map;

[0011] According to the positioning of the vehicle, local road boundary information within a preset range near the vehicle is obtained;

[0012] Constructing a point cloud data boundary corresponding to the local road boundary;

[0013] An area within the local road boundary is defined as the region of interest.

[0014] As an optional implementation, the method further includes:

[0015] defining an area outside the local road boundary as a non-interest area;

[0016] The targets in the non-interest area are divided into the static obstacle layer.

[0017] As an optional implementation, the performing target recognition and classification on the environmental point cloud data in the region of interest includes:

[0018] Preliminary identification of target type based on target motion characteristics and geometric features;

[0019] The preliminary recognition results are verified and optimized according to the regional attributes of the high-precision map.

[0020] As an optional implementation manner, the dividing the target in the region of interest into a dynamic obstacle layer or a static obstacle layer according to the motion attribute, geometric characteristics and point cloud density of the target includes:

[0021] Classifying the targets with dynamic motion attributes into the dynamic obstacle layer;

[0022] Classifying the targets with static motion attributes into the static obstacle layer;

[0023] Classify the small-sized targets whose point cloud density is lower than a preset threshold into the static obstacle layer;

[0024] Abnormal point cloud targets that fail to be detected are divided into the static obstacle layer.

[0025] As an optional implementation, the method further includes:

[0026] The original point cloud shape of the target is retained in the static obstacle layer.

[0027] As an optional implementation, the road attribute information of the high-precision map includes ramp attributes, straight road attributes and sidewalk attributes.

[0028] As an optional implementation manner, the classification and tracking results of the road attribute information optimization target based on the high-precision map include:

[0029] Based on the sidewalk attributes, verify and limit the speed of the pedestrian target;

[0030] Based on the ramp attributes, identifying and optimizing the ramp road surface;

[0031] Based on the straight road attributes, the target direction information is verified and the tracking trajectory is optimized.

[0032] As an optional implementation manner, the identifying and optimizing the slope road surface based on the ramp attribute includes:

[0033] Get the inclination angle of the ramp;

[0034] Dynamically setting a ground segmentation height threshold according to the tilt angle;

[0035] The slope road surface is segmentedly fitted based on the adjusted ground segmentation height threshold.

[0036] As an optional implementation, the method further includes:

[0037] Performing real-time obstacle avoidance path planning based on the targets in the dynamic obstacle layer;

[0038] The static obstacle layer is used to provide boundary constraint information for the path planning.

[0039] The present application provides an environmental perception method based on laser radar and high-precision map. The technical solution provided by the embodiment of the present application brings at least the following beneficial effects: the method includes: collecting environmental point cloud data in real time through laser radar; dividing the area of ​​interest based on the high-precision map; identifying and classifying the environmental point cloud data in the area of ​​interest; dividing the target in the area of ​​interest into a dynamic obstacle layer or a static obstacle layer according to the target's motion attributes, geometric characteristics and point cloud density; optimizing the classification and tracking results of the target based on the road attribute information of the high-precision map, and assisting path planning and decision-making. The present application achieves high precision and high stability of environmental perception by combining laser radar point cloud data and high-precision maps. Through the division of the area of ​​interest, the precise distinction between dynamic and static obstacle layers, and the optimization of targets based on road attributes, the computational burden is effectively reduced, the false detection rate is reduced, and the adaptability and decision-making reliability of the system in complex scenarios are improved. The method significantly enhances the efficiency and safety of obstacle avoidance path planning, supports multi-target classification and optimization, has good scalability, and is suitable for autonomous driving and other intelligent scenarios.

[0040] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 A flowchart of an environment perception method based on laser radar and high-precision map provided in an embodiment of the present application;

[0043] Figure 2 A flowchart of a method for dividing a region of interest provided in an embodiment of the present application;

[0044] Figure 3 A flowchart of a method for dividing a non-interest area provided in an embodiment of the present application;

[0045] Figure 4 A flowchart of a target recognition and classification method provided in an embodiment of the present application;

[0046] Figure 5 A flow chart of a target partitioning method provided in an embodiment of the present application;

[0047] Figure 6 A flowchart of a method for classifying optimization targets and tracking results provided in an embodiment of the present application;

[0048] Figure 7 A flowchart of a slope road surface identification and optimization method provided in an embodiment of the present application;

[0049] Figure 8 A flowchart of an environment perception application method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0051] The following will describe in detail an environment perception method based on laser radar and high-precision map provided in an embodiment of the present application in combination with a specific implementation method. Figure 1 A flowchart of an environment perception method based on laser radar and high-precision map provided in an embodiment of the present application, such as Figure 1 As shown, the specific steps are as follows:

[0052] Step 101, collect environmental point cloud data in real time through laser radar.

[0053] In practice, LiDAR acquires 3D information of objects in the environment and generates point cloud data by emitting laser pulses and receiving reflected signals. Point cloud data contains information such as the position and shape of the target object and is the basis of environmental perception. For example, in autonomous driving, a 32-line LiDAR installed on the roof or body of a vehicle collects point cloud data within a 360° range around the vehicle at a scanning frequency of 10Hz.

[0054] Step 102, dividing the area of ​​interest based on the high-precision map.

[0055] In implementation, the area around the vehicle is divided into areas of interest (such as inside the road) and areas of non-interest (such as outside the road) through the road boundary information provided by the high-precision map. The areas of interest can concentrate resources for refined processing and reduce the computational burden of irrelevant areas. For example: read the global road boundary data in the high-precision map, and combine the vehicle positioning information to extract the local road range of 50 meters in front of the vehicle and 5 meters on the left and right as the area of ​​interest. Based on the spatial distribution of point cloud data, the points in the area of ​​interest are divided into road points and non-road points.

[0056] As an optional implementation, Figure 2 A flowchart of a method for dividing a region of interest provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the specific steps of dividing the area of ​​interest based on the high-precision map in step 102 are as follows:

[0057] Step 201, read the global road boundary information in the high-precision map.

[0058] In practice, high-precision maps contain geometric information of road boundaries (such as lane lines, curbs, etc.), which are stored in the form of vector data. By parsing high-precision maps, road boundaries in the global scope can be extracted. For example: After the vehicle is started, the system connects to the high-precision map database and reads the global road boundary information of the current area, including the shape, length, width and adjacent relationship of each road. For another example, the system can extract the latitude and longitude coordinates of the road boundary and the road type (such as main road, secondary road, sidewalk, etc.) from the OSM map.

[0059] Step 202: obtaining local road boundary information within a preset range near the vehicle based on the vehicle positioning.

[0060] In implementation, the local range related to the vehicle's position is filtered out from the global road boundary in combination with the vehicle's real-time positioning information (such as GPS coordinates or IMU data). Limiting the processing range can significantly reduce the amount of calculation. For example, when the vehicle's real-time positioning shows that it is in the middle of a road section, the road boundary information within the range of 50 meters in front and behind and 5 meters in the left and right is extracted with the vehicle as the center to generate the spatial range of the local road.

[0061] Step 203: construct a point cloud data boundary corresponding to the local road boundary.

[0062] In implementation, the point cloud data collected by the laser radar can be matched with the local boundary information in the high-precision map to generate the point cloud data boundary corresponding to the road boundary, which is used to accurately divide the area of ​​interest. For example: all lanes within a preset range near the vehicle can be obtained, and the point cloud can be roughly determined whether it is within the extreme value range of a lane, and then carefully determined whether the point cloud is in the polygonal boundary point cloud box, so that the point cloud inside and outside the road can be divided. Another example: within a local range, the point cloud coordinates collected by the laser radar can be projected onto the road boundary coordinates of the high-precision map to generate a corresponding boundary point cloud. The projection algorithm can be used to overlap the laser radar point cloud with the main lane boundary, and mark the points beyond the boundary as invalid points.

[0063] Step 204: define the area within the local road boundary as an area of ​​interest.

[0064] In practice, the area within the local road boundary is usually the main range of vehicle travel and the key area of ​​environmental perception. By defining this area as the region of interest, the accuracy and efficiency of subsequent point cloud data processing and obstacle detection can be optimized. For example, all point cloud data within the local boundary can be defined as the region of interest and given a regional label, such as "point cloud within the lane". In subsequent processing, only these points are clustered and detected, such as identifying dynamic targets such as vehicles and pedestrians, while the point cloud outside the boundary is ignored.

[0065] As an optional implementation, Figure 3 A flowchart of a method for dividing a non-interest area provided in an embodiment of the present application is shown in FIG. Figure 3 As shown, the specific steps are as follows:

[0066] Step 301: define the area outside the local road boundary as a non-interest area.

[0067] In practice, the area outside the local road boundary is usually a range that vehicles cannot pass or do not need to pay attention to, such as sidewalks, buildings, etc. By defining non-interest areas, the scope and computational burden of point cloud data processing can be reduced, and efficiency and stability can be improved. For example: the system can filter the point cloud collected by the lidar based on the local boundary information of the vehicle's current road. Point clouds that exceed the road boundary (such as trees, pedestrians, or fixed buildings across the two sides of the road) are marked as non-interest area point clouds. For example, when the vehicle is on a two-lane straight road, the system will directly mark the green belt point cloud outside the road boundary as "non-interest area point cloud."

[0068] Step 302: Classify the targets in the non-interest area into a static obstacle layer.

[0069] In practice, targets in non-interest areas usually have little impact on vehicle decision-making (such as pedestrians or stationary objects outside the road). Classifying these targets into a static obstacle layer helps to avoid them occupying the drivable area or falsely triggering the obstacle avoidance mechanism. For example: the system can detect targets in non-interest areas through a point cloud clustering algorithm. For example, for trees located at the edge of the road, the system treats its complete point cloud as a static obstacle, annotates its boundary position and size, but does not perform dynamic tracking. Similarly, building point clouds in non-interest areas (such as walls, billboards, etc.) are also directly classified as static obstacles, and their three-dimensional geometric information is stored to provide scene boundary references.

[0070] Step 103: Perform target recognition and classification on the environmental point cloud data within the region of interest.

[0071] In the implementation, the geometric features of the point cloud (such as height, density, etc.) and the motion characteristics of the target (such as speed, trajectory, etc.) are used to classify the objects in the area of ​​interest, and vehicles, pedestrians and other targets are initially distinguished, providing a basis for further processing of obstacles. For example, the point cloud data can be grouped through a clustering algorithm, and the size, density and height characteristics of each group of point clouds are calculated. If the point cloud height is within the range of 0.5-2.5 meters, the density is high and the speed is close to 0, it is identified as a stationary vehicle. If the height is less than 2 meters and the speed is low, it is identified as a pedestrian target.

[0072] As an optional implementation, Figure 4 A flowchart of a target recognition and classification method provided in an embodiment of the present application is shown in FIG. Figure 4 As shown, the specific steps of performing target recognition and classification on the environmental point cloud data in the region of interest in step 103 are as follows:

[0073] Step 401, preliminarily identifying the target type based on the motion characteristics and geometric features of the target.

[0074] In practice, the target's motion characteristics (such as speed and direction) and geometric features (such as shape and size) are the main basis for distinguishing pedestrians, vehicles and stationary objects. By analyzing the motion trajectory and morphological distribution of the point cloud, the target type can be preliminarily identified, providing a basis for subsequent optimization. The system can cluster and motion analyze the point cloud in the area of ​​interest, and can use time series data to calculate the speed and direction of movement of the point cloud cluster. For example, if the point cloud cluster has a continuous moving trajectory and the speed is between 5-10km / h, it is preliminarily identified as a pedestrian. If the speed is higher than 30km / h, it may be a vehicle. The system can also determine the target type by the three-dimensional size and density of the point cloud. For example, a point cloud cluster with a length greater than 4 meters and a width of about 2 meters may be a small vehicle. A point cloud cluster with a height less than 2 meters may be a pedestrian.

[0075] Step 402, verify and optimize the preliminary recognition results according to the regional attributes of the high-precision map.

[0076] In implementation, high-precision maps provide regional attributes (such as lanes, sidewalks, intersections, etc.), and these prior information can be used to verify the point cloud target recognition results. For example, pedestrians are usually on the sidewalk, while vehicles are mostly in the lane. Combining regional attributes can correct misclassification and improve recognition accuracy. For example: the system can make a secondary judgment on the recognition results based on the regional attributes of the high-precision map. For example: if the initially identified pedestrian target is in the lane, it may be misclassified, and its shape and speed will be re-evaluated to correct the results. If a target is on a ramp, the system optimizes its motion trajectory based on the inclination information of the ramp to ensure classification stability. In the intersection area, the system further subdivides multiple targets in the area and determines whether they are vehicles or pedestrian queues through motion trajectory analysis.

[0077] Step 104 , classifying the targets in the region of interest into a dynamic obstacle layer or a static obstacle layer according to the motion attributes, geometric characteristics and point cloud density of the targets.

[0078] In implementation, dynamic targets (such as moving vehicles) are divided into dynamic obstacle layers, and stationary targets or objects of no interest (such as fences and trees) are divided into static obstacle layers by analyzing the motion state and geometric characteristics of the target. For example: among the detected targets, if the motion attributes show that the target speed is greater than 1m / s and the direction is consistent with the road, it is classified as a dynamic obstacle layer. If the target size is less than 0.3 meters or the point cloud density is low, it is directly classified into the static obstacle layer. For example, if a tree trunk is detected across the road, its height is above 0.8 meters, but the density is lower than the threshold, it is classified into the static obstacle layer.

[0079] As an optional implementation, Figure 5 A flow chart of a target partitioning method provided in an embodiment of the present application is as follows: Figure 5As shown, in step 104, the specific steps of dividing the target in the region of interest into a dynamic obstacle layer or a static obstacle layer according to the motion attributes, geometric characteristics and point cloud density of the target are as follows:

[0080] Step 501: Classify targets with dynamic motion attributes into a dynamic obstacle layer.

[0081] In practice, dynamic targets are obstacles that have a greater impact on vehicle driving, such as pedestrians and vehicles. Through continuous frame motion analysis of point clouds, its motion properties can be identified, and moving targets can be divided into dynamic obstacle layers for subsequent tracking and obstacle avoidance planning. For example: the time series point cloud data of the lidar can be used to analyze the continuous position changes of the point cloud cluster. If a point cloud cluster has a significant speed (such as ≥1m / s), it is determined to be a dynamic target, which may be a moving pedestrian or vehicle. Targets with dynamic motion characteristics (such as by detecting moving pedestrians, vehicles or riding bicycles) are stored in the dynamic obstacle layer.

[0082] Step 502: Classify the targets with static motion attributes into the static obstacle layer.

[0083] In practice, static targets such as buildings, street lights, or parked vehicles will not cause real-time interference to dynamic obstacle avoidance. Classifying point cloud targets with static motion attributes into a static obstacle layer can effectively reduce the computational burden of dynamic obstacle avoidance. For example, the system can perform motion trajectory analysis on a point cloud cluster. If the point cloud target does not move in multiple frames and the speed is less than 0.1m / s, it is judged as a static target, such as a parked vehicle or a fixed roadblock. The system can classify such targets into a static obstacle layer and record their positions to assist in path boundary planning.

[0084] Step 503: Classify small-sized targets whose point cloud density is lower than a preset threshold into a static obstacle layer.

[0085] In practice, small-sized targets with low point cloud density are usually misidentified or irrelevant interferences (such as leaves, gravel, etc.). Classifying them into the static obstacle layer can avoid affecting the accuracy of path planning. For example, the system can calculate the point cloud density and size for each point cloud cluster. If the point cloud density of the target is lower than the set threshold (such as 10 points / m 3 ) and the size is less than 0.5m 3 , the system determines it as a non-threatening object (such as a small animal or garbage). The target is classified into the static obstacle layer and marked as a low priority obstacle.

[0086] Step 504: Classify the abnormal point cloud targets that fail to be detected into the static obstacle layer.

[0087] In implementation, abnormal point cloud targets may fail to be classified due to data loss, noise or irregular shapes. Classifying these targets into the static obstacle layer can avoid the interference of incorrect classification on dynamic obstacle avoidance. For example: for point clouds that fail to complete clustering or detection, if a target point cloud has no clear shape or motion trajectory, the system can determine it as an abnormal target (such as complex tree crowns, low-quality point clouds). The system can classify it into the static obstacle layer and only retain its original shape and position to assist in environment modeling. For example, the point cloud of branches across the road can be classified into the static obstacle layer to avoid erroneous triggering of the obstacle avoidance mechanism.

[0088] As an optional implementation, the original point cloud shape of the target is retained in the static obstacle layer.

[0089] In implementation, retaining the original point cloud shape of static obstacles helps to accurately represent the spatial characteristics of the target, thereby improving the accuracy of environmental modeling, especially in scenes with irregular obstacles (such as tree crowns, billboards or special-shaped roadblocks), which can avoid the simplified polygon envelope from occupying too much drivable area and optimize path planning and decision-making effects. For example, the system can mark the point cloud clusters in the static obstacle layer and distinguish them into targets such as buildings, trees, and parked vehicles. For example, when a roadside tree crown is detected, the point cloud is irregularly distributed and cannot be accurately fitted by the geometric model. No envelope approximation is performed on the static target, and only the complete point cloud data is stored in the static obstacle layer. For the above tree crown, all its point cloud coordinates are retained, and the height and coverage of the lowest point are recorded. When the system is planning the path, the dynamic obstacles will be updated in real time, and the point cloud data of the static obstacle layer can be directly used to define the boundary of the obstacle avoidance area. When the vehicle approaches the tree crown, the system can judge whether it affects the vehicle's passage based on the height of the lowest point of the tree crown. If it is lower than the vehicle height limit, it is marked as an area to be avoided, otherwise it is only used as reference background data.

[0090] Step 105, optimizing the classification and tracking results of the target based on the road attribute information of the high-precision map, and assisting in path planning and decision-making.

[0091] In implementation, road attributes in high-precision maps (such as ramps, straight roads, sidewalks, etc.) provide prior information for target classification and are used to correct tracking results and optimize the vehicle's path planning and obstacle avoidance strategies. For example, straight road information can be used to filter target trajectory jitter, and ramp information can improve the accuracy of ground segmentation. For example: the speed of pedestrian targets can be limited in combination with sidewalk attributes. If the target speed is detected to be too high and the trajectory deviates from the sidewalk, it is corrected to a non-pedestrian target. If the ramp information shows that the current road is tilted 10 degrees, the height threshold of the ground segmentation is dynamically adjusted to 0.3 meters to avoid misjudging the ramp as an obstacle. At the same time, avoid targets in the static obstacle layer such as roadside railings when planning the path.

[0092] As an optional implementation, Figure 6 A flowchart of a method for classifying optimization targets and tracking results provided in an embodiment of the present application, such as Figure 6 As shown, the road attribute information of the high-precision map includes ramp attributes, straight road attributes and sidewalk attributes. The specific steps of classifying and tracking the results of the optimization target based on the road attribute information of the high-precision map in step 105 are as follows:

[0093] Step 601, based on the sidewalk attributes, the speed of the pedestrian target is verified and limited.

[0094] In implementation, the sidewalk attributes provide a priori constraints for target classification. By defining the range of the sidewalk in the high-precision map, the speed of pedestrian targets can be limited to avoid misidentifying slow-moving objects (such as carts or stagnant targets) as dynamic pedestrians, and reduce erroneous obstacle avoidance or path planning triggers. For example: the system detects a moving target on the sidewalk at a speed of 8km / h. According to the speed limit of the sidewalk attribute (such as 10km / h), the target is verified as a pedestrian. If the speed exceeds the normal value of the sidewalk (such as 30km / h), it is reclassified as a non-pedestrian target, which may be a cyclist or a vehicle. In path planning, interactions with pedestrian targets are restricted, for example: when the vehicle approaches the sidewalk, set the speed condition for the target pedestrian to give priority to avoidance. If multiple target pedestrians are detected on the sidewalk, the system adjusts the planned path, slows down to pass, or chooses to bypass the non-interfering area.

[0095] Step 602: Based on the ramp attributes, the ramp road surface is identified and optimized.

[0096] In implementation, the special terrain in the ramp area easily leads to ground mis-segmentation and target misclassification. For example, the LiDAR point cloud may have a height deviation due to the slope of the ramp, causing the ground to be misidentified as an obstacle. Optimize ground recognition through ramp attributes to improve segmentation accuracy. For example: the system can adjust the ground segmentation height threshold of the LiDAR point cloud according to the ramp inclination information in the high-precision map (such as a 10° slope), and include the slope point cloud in the ground classification. For example, the ground threshold of ordinary roads is 0.2m, and it is increased to 0.5m on the ramp to adapt to the slope changes. When the self-driving car enters the ramp, the ramp road surface model is refitted to filter out non-ground targets (such as parked vehicles or roadside guardrails).

[0097] Step 603: based on the straight road attribute, verify the target direction information and optimize the tracking trajectory.

[0098] In implementation, the straight road attribute provides directional constraints for target tracking, which can reduce the trajectory jitter problem caused by sensor errors or environmental interference. For example, in a straight road scenario, the target's movement direction should be consistent with the road direction, and the direction change beyond the range can be regarded as an abnormal trajectory and corrected. For example: On a straight road, a vehicle target is detected to be moving in a direction of 30°, while the straight road direction is 0°. The system determines that its trajectory is abnormal and corrects it to the road direction. If the target direction deviates too much (such as 90°), it is marked as a misclassified or non-relevant target. Directional constraints are imposed on the historical motion trajectory of the target. For example, the trajectory of a vehicle on a straight road should be a smooth straight line. If the trajectory shows unreasonable left and right swings, it is adjusted to a stable path through a filtering algorithm. When the ego vehicle is tracking the preceding vehicle on a straight road, the optimized target trajectory can improve the path stability of the ego vehicle and avoid unnecessary braking or acceleration caused by target trajectory jitter.

[0099] As an optional implementation, Figure 7 A flowchart of a slope road surface recognition optimization method provided in an embodiment of the present application is shown in FIG. Figure 7 As shown, the specific steps of identifying and optimizing the ramp road surface based on the ramp attributes in step 602 are as follows:

[0100] Step 701, obtaining the inclination angle of the ramp.

[0101] In implementation, the ramp attributes include the slope angle of the ramp as the core parameter to describe the terrain characteristics. By reading the slope angle, the system can adjust the processing algorithm of the point cloud data to adapt to the special terrain of the ramp section and avoid misjudgment of ground segmentation. For example: the system can read the road attributes of the current position of the vehicle through the high-precision map interface to obtain the slope angle of the ramp area, such as a 10° uphill slope. If the map contains the starting and ending coordinates of the ramp, the specific slope section can also be further confirmed in combination with the positioning of the vehicle. When the vehicle enters a new section, the slope angle is updated in real time. For example, the current ramp is 10° uphill, and after driving for a while, it enters a 15° steep slope. The system dynamically adjusts the relevant parameters.

[0102] Step 702: dynamically set the ground segmentation height threshold according to the tilt angle.

[0103] In implementation, the inclination of the ramp can cause the point cloud data of the lidar to shift in the height direction. By adjusting the height threshold of the ground segmentation, the ground and non-ground targets (such as vehicles, guardrails, etc.) can be accurately distinguished to avoid the problem of ground mis-segmentation or missed segmentation. The system can dynamically adjust the threshold according to the inclination angle of the ramp. For example: when the inclination angle of the ramp is 0°, the ground height threshold is 0.2m. When the inclination angle of the ramp is 10°, the threshold increases to 0.4m. When the inclination angle reaches 20°, the threshold is set to 0.6m. The system detects that the inclination angle of the ramp is 12° and dynamically sets the ground height threshold to 0.45m. Points in the lidar point cloud with a height below 0.45m are identified as the ground to avoid mistakenly marking the outer edge of the slope as the ground area.

[0104] Step 703: Perform segmented fitting on the slope road surface based on the adjusted ground segmentation height threshold.

[0105] In practice, in ramp sections, a single plane model cannot accurately describe the complex ground morphology. By fitting the ground in sections, the system can generate a more accurate ramp road surface model, improve perception accuracy, and provide reliable support for subsequent path planning. The system can divide the ramp road surface into several small sections, and fit a plane in each section based on local point cloud data. For example: the ramp is divided into uniform small sections of 5 meters, and each section is fitted with an inclined plane. By splicing them section by section, a complete ramp road surface model is generated. If a segment of the point cloud is found to deviate from the overall trend (such as an abrupt step) during the fitting process, the system can automatically adjust the fitting parameters of the segment to make it smoothly connected with the adjacent segments. The system can also verify the fitting results to ensure that they match the ramp attributes of the high-precision map, such as the starting point, end point, and inclination angle.

[0106] As an optional implementation, Figure 8 A flowchart of an environment perception application method provided in an embodiment of the present application is shown in FIG. Figure 8 As shown, the application of dynamic obstacle layer and static obstacle layer is as follows:

[0107] Step 801, performing real-time obstacle avoidance path planning based on targets in the dynamic obstacle layer.

[0108] In practice, the targets in the dynamic obstacle layer are moving objects, such as vehicles or pedestrians, whose trajectories are uncertain. By tracking these targets in real time, the system can predict their movement trends and dynamically adjust the obstacle avoidance path to ensure driving safety and smoothness. Specifically, the trajectory of the targets in the dynamic obstacle layer is predicted, for example: pedestrians move along the sidewalk at a speed of 1.5m / s, and the vehicle in front is driving in a straight line at a speed of 10m / s, and there is a possibility of changing lanes. Based on the movement trend of the target, the system adds dynamic obstacle avoidance areas to the self-vehicle path planning to avoid collisions. The system can also generate obstacle avoidance paths based on the real-time position and movement trajectory of dynamic obstacles. For example: there is a slow-moving vehicle 10 meters ahead, and the system plans a path to change lanes to the left. If the path cannot meet the safety distance requirements, the system reduces the speed of the self-vehicle to avoid collisions. When the movement trajectory of the dynamic obstacle changes (such as sudden stopping or turning), the system updates the obstacle avoidance path in real time. For example, if a pedestrian suddenly accelerates to cross the road on a zebra crossing, the self-vehicle will immediately adjust the deceleration or parking strategy.

[0109] Step 802: Use the static obstacle layer to provide boundary constraint information for path planning.

[0110] In practice, the targets in the static obstacle layer are fixed objects (such as guardrails, buildings, or road signs) whose positions are relatively stable. By providing the boundary information of static obstacles for path planning, the system can ensure that path planning is carried out within a reasonable range to avoid collisions or driving off the road. Specifically, the system can extract the location and contour data of obstacles from the static obstacle layer. For example, the guardrail on the right side of the road is 1.5 meters away from the vehicle, the width of the static roadblock is 0.8 meters, and it is located 1 meter to the left of the center of the lane. In path planning, the boundaries of static obstacles are set as inaccessible areas. For example, the left boundary of the planned path of the vehicle is the lane line, and the right boundary is the outer edge of the guardrail. Ensure that the vehicle always drives between the two boundaries and does not deviate from the safe area. For example, when the vehicle is driving on a narrow road, the system can use the static obstacle layer to ensure that the path planning is close to the boundary while avoiding collisions. When the system detects a construction fence in front, it automatically adjusts the planned path so that the vehicle detours and returns to the safe driving area.

[0111] The embodiment of the present application provides an environmental perception method based on laser radar and high-precision map, the method comprising: collecting environmental point cloud data in real time through laser radar. Dividing the region of interest based on the high-precision map. Performing target recognition and classification on the environmental point cloud data in the region of interest. According to the motion attributes, geometric characteristics and point cloud density of the target, the target in the region of interest is divided into a dynamic obstacle layer or a static obstacle layer. The classification and tracking results of the target are optimized based on the road attribute information of the high-precision map, and the path planning and decision are assisted. The embodiment of the present application combines the laser radar point cloud data and the high-precision map information, and the method realizes accurate perception of the environment, reduces the noise interference in the traditional point cloud processing process, and maintains a high stability in complex scenes (such as ramps and intersections). Through the division of the region of interest, only the point cloud within the road boundary is clustered and detected, avoiding redundant calculation of data in non-critical areas, and improving the calculation efficiency. Based on the motion attributes, geometric characteristics and point cloud density of the target, the obstacle type is dynamically distinguished, so that the drivable area occupied by static obstacles is minimized, and the dynamic obstacle detection is more accurate, providing more reliable information for vehicle decision-making. By utilizing the road attribute information (such as ramps, straight roads, and sidewalks) of high-precision maps, the target tracking and ground segmentation algorithms are effectively optimized, and the applicability and accuracy of the system in special scenarios (such as ramp fitting or straight target direction correction) are improved. The dynamic obstacle layer provides real-time dynamic information for path planning, while the static obstacle layer defines boundary constraints by retaining the point cloud shape, avoiding misjudgment of the driving area, and improving the feasibility and safety of the obstacle avoidance path. Combining the characteristics of the target shape, speed, and high-precision map attributes, the target classification algorithm is optimized, especially in the absence of visual assistance. Effective distinction between vehicles and pedestrians is achieved, and the intelligent level of environmental perception is improved. In scenarios such as tree crowns occupying the road, irregular objects crossing the border, or physically isolated targets (such as elevated layered roads), the division strategy of the static obstacle layer effectively avoids misjudgment and unnecessary obstacle avoidance actions, and improves the stability and safety of vehicle driving. The method of the embodiment of the present application can not only be applied to autonomous driving scenarios, but can also be extended to unmanned delivery vehicles, security monitoring and other fields, providing an effective solution for environmental perception in multiple scenarios.

[0112] It should be understood that although Figures 1 to 8 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figures 1 to 8At least part of the steps may include multiple steps or multiple stages. These steps or stages are not necessarily performed at the same time, but can be performed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed in turn or alternately with other steps or at least part of the steps or stages in other steps.

[0113] It can be understood that the same / similar parts between the various embodiments of the above method in this specification can refer to each other, and each embodiment focuses on the differences from other embodiments. For related points, please refer to the description of other method embodiments.

[0114] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0115] It should also be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data for analysis, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0116] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0117] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0118] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. An environment perception method based on laser radar and high-precision map, characterized in that: The method comprises: Collect environmental point cloud data in real time through LiDAR; Divide the area of ​​interest based on high-precision maps; Performing target recognition and classification on the environmental point cloud data within the region of interest; According to the motion attributes, geometric characteristics and point cloud density of the target, the target in the region of interest is divided into a dynamic obstacle layer or a static obstacle layer; The classification and tracking results of the target are optimized based on the road attribute information of the high-precision map, and path planning and decision-making are assisted.

2. The method according to claim 1, characterized in that The dividing of the area of ​​interest based on the high-precision map includes: Reading global road boundary information in the high-precision map; According to the positioning of the vehicle, local road boundary information within a preset range near the vehicle is obtained; Constructing a point cloud data boundary corresponding to the local road boundary; An area within the local road boundary is defined as the region of interest.

3. The method according to claim 2, characterized in that The method further comprises: defining an area outside the local road boundary as a non-interest area; The targets in the non-interest area are divided into the static obstacle layer.

4. The method according to claim 1, characterized in that: The target recognition and classification of the environmental point cloud data in the region of interest includes: Preliminary identification of target type based on target motion characteristics and geometric features; The preliminary recognition results are verified and optimized according to the regional attributes of the high-precision map.

5. The method according to claim 1, characterized in that The step of dividing the target in the region of interest into a dynamic obstacle layer or a static obstacle layer according to the motion attributes, geometric characteristics and point cloud density of the target includes: Classifying the targets with dynamic motion attributes into the dynamic obstacle layer; Classifying the targets with static motion attributes into the static obstacle layer; Classify the small-sized targets whose point cloud density is lower than a preset threshold into the static obstacle layer; Abnormal point cloud targets that fail to be detected are divided into the static obstacle layer.

6. The method according to claim 5, characterized in that The method further comprises: The original point cloud shape of the target is retained in the static obstacle layer.

7. The method according to claim 1, characterized in that The road attribute information of the high-precision map includes ramp attributes, straight road attributes and sidewalk attributes.

8. The method according to claim 7, characterized in that The classification and tracking results of the road attribute information optimization target based on the high-precision map include: Based on the sidewalk attributes, verify and limit the speed of the pedestrian target; Based on the ramp attributes, identifying and optimizing the ramp road surface; Based on the straight road attributes, the target direction information is verified and the tracking trajectory is optimized.

9. The method according to claim 8, characterized in that The step of identifying and optimizing the slope road surface based on the slope attribute includes: Get the inclination angle of the ramp; Dynamically setting a ground segmentation height threshold according to the tilt angle; The slope road surface is segmentedly fitted based on the adjusted ground segmentation height threshold.

10. The method according to claim 1, characterized in that The method further comprises: Performing real-time obstacle avoidance path planning based on the targets in the dynamic obstacle layer; The static obstacle layer is used to provide boundary constraint information for the path planning.

Images