A travelable area determination method and apparatus

Abstract

The application provides a drivable area determination method and device; comprising: obtaining multi-source sensor data and vehicle driving environment; performing non-uniform polar coordinate sector grid projection and data structuring processing on the multi-source sensor data to obtain first grid information corresponding to each grid; determining conflict coding features corresponding to each type of spatial point in the grid according to the vehicle driving environment and the first grid information; obtaining reference logarithmic confidence corresponding to each grid, and obtaining target logarithmic confidence of the grid according to the conflict coding features corresponding to each type of spatial point in each grid, the first grid information and the reference logarithmic confidence; for the same sector, traversing the target logarithmic confidence of each grid from near to far along the radial direction of the sector, determining the polar radius of the grid with the first target logarithmic confidence meeting the confidence condition as the target polar radius of the sector, and determining the current drivable area of the vehicle according to the target polar radius corresponding to all sectors.

Description

Technical Field

[0001] This application relates to the field of intelligent driving technology for vehicles, and in particular to a method and device for determining a drivable area. Background Technology

[0002] With the continuous development of artificial intelligence technology, intelligent driving systems have gradually acquired various practical functions such as automatic parking, automatic emergency braking, and lane departure warning. During intelligent driving, vehicles use multi-sensor fusion perception to detect and analyze the surrounding environment in real time, thereby identifying potential risks and effectively reducing the accident rate through proactive intervention or timely warnings, providing reliable support for the realization of the above functions. It is evident that accurate and reliable drivable area information is a key prerequisite for ensuring the safety and reliability of intelligent driving decisions. Therefore, how to detect drivable areas to obtain more accurate and reliable drivable area information has become a particularly important part of intelligent vehicles. Summary of the Invention

[0003] One objective of this application is to provide a method for determining a drivable area; another objective is to provide a device for determining a drivable area; a third objective is to provide an apparatus for determining a drivable area; a fourth objective is to provide a vehicle; a fifth objective is to provide a computer-readable storage medium; and a sixth objective is to provide a computer program product.

[0004] To achieve the above objectives, the technical solution of this application embodiment is implemented as follows: This application provides a method for determining a drivable area, the method including: The system acquires multi-source sensor data and the vehicle's driving environment around the vehicle. The multi-source sensor data includes: visual spatial point cloud and / or target attribute information output by the visual sensor, and radar spatial point cloud output by the radar sensor. The spatial points are the boundary points of the drivable area identified by the sensors. Multi-source sensor data is subjected to non-uniform polar coordinate sector grid projection and data structuring to obtain the first grid information corresponding to the grid to which various spatial points belong; wherein, the first grid information includes at least one of the following: the polar radius of various spatial points under the non-uniform polar coordinate grid, sector identifier, distance band identifier under the sector corresponding to the sector identifier, sensor source identifier, first acquisition time, target attribute information, sensor basic reliability and historical stability count; Based on the vehicle driving environment and the first grid information corresponding to each grid, determine the conflict coding features corresponding to various spatial points within the grid. Obtain the baseline log confidence score for each grid. Based on the conflict coding features of various types of spatial points in each grid, the first grid information, and the baseline log confidence score, traverse all types of spatial points in the grid to perform multi-source data fusion and obtain the target log confidence score of the grid. For the same sector, the target log confidence of each grid is traversed radially from near to far. The polar radius of the first grid that meets the target log confidence condition is determined as the target polar radius of the sector. Based on the target polar radii of all sectors, the current drivable area of ​​the vehicle is determined.

[0005] Based on the aforementioned technical means, visual and radar point clouds are uniformly mapped to the same frame through non-uniform polar coordinate grid projection, recording multi-dimensional grid information such as polar radius, sector, range band, source, time, attributes, reliability, and historical counts, thus solving the problems of heterogeneous data scale inconsistency and time sequence alignment. Conflict coding features are dynamically generated based on the driving environment, enabling the system to intelligently distinguish between real physical conflicts and environmental perception differences, avoiding misjudgments in scenarios such as tunnel walls and guardrails. A benchmark logarithmic confidence level is introduced, and conflict coding, sensor reliability, and historical stable counts are fused for grid-by-grid iterative updates, suppressing instantaneous noise and perception degradation interference, and improving confidence robustness. Finally, the first grid polar radius that meets the confidence condition is searched radially along the sector as the target boundary, and all sector target polar radii constitute the drivable area. Complex boundary detection is transformed into a one-dimensional threshold search, with low computational load, high real-time performance, and output of continuous, smooth, physically reliable boundaries, avoiding abrupt changes or holes, thereby obtaining a high-precision, highly robust drivable area.

[0006] This application provides a drivable area determination device, the device comprising: The acquisition module is used to acquire multi-source sensor data around the vehicle and the vehicle's driving environment; wherein, the multi-source sensor data includes: visual spatial point cloud and / or target attribute information output by the visual sensor, and radar spatial point cloud output by the radar sensor, where the spatial points are the boundary points of the drivable area identified by the sensors. The processing module is used to perform non-uniform polar coordinate sector grid projection and data structuring on multi-source sensor data to obtain the first grid information corresponding to the grid to which various types of spatial points belong; wherein, the first grid information includes at least one of the following: the polar radius of various types of spatial points under the non-uniform polar coordinate grid, sector identifier, distance band identifier under the sector corresponding to the sector identifier, sensor source identifier, first acquisition time, target attribute information, sensor basic reliability and historical stability count; The determination module is used to determine the conflict coding features corresponding to various spatial points within a grid based on the vehicle driving environment and the first grid information corresponding to each grid. The module is also used to obtain the baseline log confidence level for each grid cell; The processing module is also used to traverse all types of spatial points in the grid to perform multi-source data fusion based on the conflict coding features, first grid information and baseline log confidence of each type of spatial point in the grid, and to obtain the target log confidence of the grid. The determination module is also used to traverse the target log confidence of each grid along the radial direction of the same sector from near to far, determine the polar radius of the first grid that meets the target log confidence condition as the target polar radius of the sector, and determine the current drivable area of ​​the vehicle based on the target polar radii of all sectors.

[0007] This application provides a device for determining a drivable area, including: Memory is used to store executable instructions or computer programs. The processor, when executing computer-executable instructions or computer programs stored in the memory, implements the drivable area determination method provided in the embodiments of this application.

[0008] This application provides a vehicle that includes the aforementioned drivable area determination device.

[0009] This application provides a computer-readable storage medium storing a computer program or computer-executable instructions, which, when executed by a processor, implements the drivable area determination method provided in this application.

[0010] This application provides a computer program product, including a computer program or computer executable instructions. When the computer program or computer executable instructions are executed by a processor, the method for determining the drivable area provided in this application is implemented. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating the method for determining the drivable area provided in the embodiments of this application. Figure 1 ; Figure 2 This is a flowchart illustrating the multi-sensor freespace point fusion method for a low-computing-power platform provided in this application embodiment. Figure 2 ; Figure 3 A schematic diagram of the drivable area determination device provided in this application embodiment; Figure 4 This is a schematic diagram of the drivable area determination device provided in the embodiments of this application; Figure 5 This is a schematic diagram of the vehicle structure provided in the embodiments of this application.

[0012] It should be noted that the terms "first" and "second" mentioned above are only used to distinguish between different options and do not represent the degree of superiority or inferiority of the options or their priority in the implementation process. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0014] This application provides a method for determining a drivable area, applied to a drivable area determination device. This method enables accurate detection of the drivable area surrounding a vehicle. The drivable area determination device may include a mobile phone, tablet computer, laptop computer, PDA, in-vehicle electronic device, wearable device, or ultra-mobile personal computer. Mobile personal computers (UMPCs), netbooks or personal digital assistants (PDAs), servers, network attached storage (NAS), personal computers (PCs), etc.

[0015] Reference Figure 1 As shown, Figure 1 A schematic diagram of the implementation process of a drivable area determination method provided in this application embodiment. Figure 1 The method may include the following steps: Step 101: Obtain multi-source sensor data around the vehicle and the vehicle's driving environment.

[0016] The multi-source sensor data includes: visual spatial point cloud and / or target attribute information output by the visual sensor, and radar spatial point cloud output by the radar sensor. The spatial points are the boundary points of the drivable area identified by the sensors.

[0017] It can be understood that spatial points, also known as freespace points, are the boundary points of the drivable area (also known as the passable area) of a vehicle in that direction. Visual spatial point clouds are the spatial point clouds output by visual sensors deployed on the vehicle. Radar spatial point clouds are the spatial point clouds acquired by radar sensors deployed on the vehicle.

[0018] In this embodiment, the visual sensor is used to collect visual image data of the area surrounding the vehicle. The information output by the visual sensor includes not only the planar coordinates of each visual spatial point in the visual spatial point cloud in the Cartesian coordinate system, but also target attribute information of each visual spatial point in the visual spatial point cloud. The target attribute information includes object type, dynamic / static attributes, and occlusion attributes. Object types include, but are not limited to, pedestrians, curbs, vehicles, road surfaces, cyclists, and signs. Dynamic / static attributes include static, dynamic, or motion attributes. Occlusion attributes include, but are not limited to, no occlusion, partial occlusion, and severe occlusion. Here, object type, dynamic / static attributes, and occlusion attributes can be represented by bitmasks. For example, Bits 0-2 represent object type (0=unknown, 1=road surface, 2=curb, 3=vehicle, 4=pedestrian, 5=cyclist), Bit 3 represents dynamic attributes (0=static, 1=dynamic / motion), and Bits 4-5 represent occlusion attributes (0=no occlusion, 1=partial occlusion, 2=severe occlusion).

[0019] In some embodiments, the information output by the visual sensor may further include the acquisition time of the visual spatial point cloud and image quality indicators. The image quality indicators are used to calculate the sensor's basic reliability. These indicators can be categorized from dimensions such as lighting environment, weather environment, special scene, image state, road type, and vehicle condition. Specifically, the lighting environment dimension includes daytime and nighttime indicators; the weather environment dimension includes sunny, cloudy, and rainy / foggy indicators; the special scene dimension includes non-tunnel and tunnel indicators; the image state dimension includes a blurred indicator; the road type dimension includes highway, urban road, and rural / off-road road indicators; and the vehicle condition dimension includes stationary, low-speed, medium-speed, and high-speed indicators. It should be noted that different indicators within the same dimension correspond to different factor coefficients used to calculate the sensor's basic reliability; a higher coefficient indicates higher sensor basic reliability. For example, the factor coefficient is 1.0 for daytime and 0.6 for nighttime; 1.0 for non-tunnel and 0.7 for tunnel; 1.0 for sunny days, 0.85 for cloudy days, and 0.45 for rainy / foggy days. Other dimensions are similar, and examples will not be provided here.

[0020] In this embodiment, the radar sensor may include a forward-facing millimeter-wave radar sensor and an angular radar sensor. The forward-facing millimeter-wave radar sensor is used to collect radar data in the forward region directly in front of the vehicle, and the angular radar sensor is used to collect radar data in the left and right lateral regions of the vehicle. The radar spatial point cloud output by the radar sensor includes the forward-facing radar spatial point cloud output by the forward-facing millimeter-wave radar sensor and the angular radar spatial point cloud output by the angular radar sensor.

[0021] In this embodiment, the coordinates of radar spatial points in the radar spatial point cloud output by the radar sensor can be polar coordinates in a polar coordinate system. It should be noted that since the forward millimeter-wave radar sensor and the angular radar sensor collect data from different areas around the vehicle, the polar coordinate system of the forward radar spatial point cloud output by the forward millimeter-wave radar sensor can also be different from the polar coordinate system of the angular radar spatial point cloud output by the angular radar sensor. For example, the forward region directly in front of the vehicle can be a sector-shaped region with the vehicle as the origin of the polar coordinate system and a first preset angular resolution of 0.5 degrees. For the left and right lateral regions of the vehicle, they can be sector-shaped regions with the vehicle as the origin of the polar coordinate system and a second preset angular resolution of 0.5 degrees. The second preset angular resolution is greater than the first preset angular resolution and is an integer multiple of the first preset angular resolution. For example, the left and right lateral regions of the vehicle can be divided into multiple sector-shaped regions with the vehicle as the origin of the polar coordinate system and a second preset angular resolution of 1 degree. Thus, based on the non-uniform polar coordinate grid angle partitioning of the radar sensor angular resolution, the sector width of different angle intervals is set differently according to the radar detection accuracy, adapting to the radar's near and far field detection characteristics.

[0022] In this embodiment, the vehicle driving environment can be determined by acquiring landmarks in the surrounding environment of the vehicle using visual sensors deployed on the target vehicle. In one possible implementation, the drivable area determination device stores map information, and the vehicle driving environment can also be determined based on the vehicle's location information and the map information. The vehicle driving environment includes, but is not limited to, scenarios such as tunnels, bridges, and highways.

[0023] Step 102: Perform non-uniform polar coordinate sector grid projection and data structuring on the multi-source sensor data to obtain the first grid information corresponding to the grid to which various spatial points belong.

[0024] The first grid information includes at least one of the following: the polar radius of various spatial points in a non-uniform polar coordinate grid, sector identifier, distance band identifier of the sector corresponding to the sector to which the sector belongs, sensor source identifier, first acquisition time, target attribute information, sensor basic reliability and historical stability count.

[0025] In this embodiment, since the forward millimeter-wave radar sensor collects radar data of the forward area directly in front of the vehicle, and the corner radar sensor collects radar data of the left and right side areas of the vehicle, and the preset angular resolutions of the two types of radar sensors are different, a non-uniform sparse polar coordinate grid that is aligned with the radar angular resolution is constructed based on the forward millimeter-wave radar sensor and the corner radar sensor, i.e., a non-uniform polar coordinate sector grid. This non-uniform sparse polar coordinate grid can be a sector grid that is dense in the middle and sparse on both sides.

[0026] In this embodiment, the multi-source sensor data includes visual spatial point clouds and radar spatial point clouds. Since the output format of the radar spatial point cloud is polar coordinates, the polar coordinates of each radar spatial point in the forward radar spatial point cloud are directly mapped to a grid with the angular resolution corresponding to the forward millimeter-wave radar sensor, and the polar coordinates of each radar spatial point in the angular radar spatial point cloud are directly mapped to a grid with the angular resolution corresponding to the angular millimeter-wave radar sensor. The visual spatial point cloud needs to be projected into a non-uniform polar coordinate sector grid based on the area where each visual spatial point is located (the forward area directly in front of the vehicle or the left and right lateral areas of the vehicle) to obtain the coordinates in the non-uniform polar coordinate system. After the multi-source sensor data projection mapping is completed, the first grid information of the grid to which various types of spatial points belong is obtained. In this way, it fully matches the physical detection capability of the hardware sensors, avoids artificially creating non-existent observation points (also known as spatial points), and reduces the number of invalid lateral subdivision grids, halving the storage and computation load.

[0027] In this embodiment of the application, the first grid information may be structure information, and the first grid information Ev may include at least one of the following: the polar radius r of various spatial points (such as visual spatial points and radar spatial points) in a non-uniform polar coordinate grid, the sector identifier theta, the range band identifier dista under the sector corresponding to the sector identifier, the sensor source identifier src, the first acquisition time t, the target attribute information mask, the sensor basic reliability q, the historical stability count hist, and the high-altitude identifier flag.

[0028] In one possible implementation, the first grid information corresponding to the grid to which each visual spatial point belongs in the visual spatial point cloud may include: the polar radius r of the visual spatial point in a non-uniform polar coordinate grid, the sector identifier theta, the distance band identifier dista of the sector corresponding to the sector to which the sector identifier belongs, the sensor source identifier src, the first acquisition time t, the target attribute information mask, and the sensor basic reliability q.

[0029] In another possible implementation, the first grid information corresponding to the grid to which each radar spatial point belongs in the radar spatial point cloud may include: the polar radius r of the radar spatial point in a non-uniform polar coordinate grid, the sector identifier theta, the range band identifier dista of the sector corresponding to the sector to which the sector identifier belongs, the sensor source identifier src, the first acquisition time t, the historical stability count hist, and the high-altitude identifier flag.

[0030] In some embodiments, since the visual spatial points in the visual spatial point cloud are located in the Cartesian coordinate system, it is necessary to perform non-uniform polar coordinate grid projection on the points in the visual spatial point cloud to map them to the non-uniform polar coordinate system. Here, performing non-uniform polar coordinate sector grid projection and data structuring on multi-source sensor data to obtain the first grid information corresponding to the grid to which various types of spatial points belong can also be achieved through the following steps: according to the angular resolution corresponding to various radar sensors, perform non-uniform polar coordinate grid projection on the visual spatial point cloud according to the preset projection rules to determine the sector identifier of the grid to which each visual spatial point belongs, and the range band identifier under the sector corresponding to the sector identifier; perform unified structuring on the visual spatial points according to the sector identifier and range band identifier of the grid to which the visual spatial points belong to, to obtain the first grid information corresponding to the grid to which the visual spatial points belong.

[0031] In this embodiment of the application, the preset projection rules include a first mapping table and a corresponding second mapping table. The first mapping table can be constructed according to the angular resolution of the forward radar sensor and is used to store the corresponding mapping relationship between the Cartesian coordinate system plane coordinates and the first polar coordinate system polar coordinates. The second mapping table can be constructed according to the angular resolution of the angular radar sensor and is used to store the corresponding mapping relationship between the Cartesian coordinate system plane coordinates and the second polar coordinate system polar coordinates.

[0032] In this embodiment, a first sub-visual spatial point cloud located in the forward region directly in front of the vehicle and a second sub-visual spatial point cloud located in the left and right lateral regions of the vehicle can be obtained from the visual spatial point cloud. In a first mapping table, the polar coordinates corresponding to the planar coordinates of each visual spatial point in the first sub-visual spatial point cloud are searched, thereby determining the sector identifier of the grid to which each visual spatial point belongs, and the range band identifier under the corresponding sector, based on these polar coordinates. Similarly, in a second mapping table, the polar coordinates corresponding to the planar coordinates of each visual spatial point in the second sub-visual spatial point cloud are searched, thereby determining the sector identifier of the grid to which each visual spatial point belongs, and the range band identifier under the corresponding sector, based on these polar coordinates. Thus, by using various radar angular resolutions to achieve non-uniform polar coordinate grid projection of the visual point cloud, and combining sector and range band identifiers to partition and classify the point cloud and output grid information in a structured manner, the grid division matches radar detection characteristics, achieving balanced near-field and far-field perception accuracy. Furthermore, the unified point cloud data structure simplifies the multi-sensor fusion preprocessing process, improves fusion computation efficiency, and enhances the adaptability and versatility for multiple radar scenarios.

[0033] Step 103: Based on the vehicle driving environment and the first grid information corresponding to each grid, determine the conflict coding features corresponding to each type of spatial point within the grid.

[0034] In this application embodiment, the conflict coding features include at least one of the following: consistency features, visual close-range closure-radar long-range passability features, radar close-range blocking-visual long-range passability features, tunnel wall suspected features, high-altitude target suspected features, time expiration features, and historical jump features.

[0035] In this embodiment of the application, the conflict coding features corresponding to the visual spatial points in the grid are determined based on the vehicle driving environment and the first grid information corresponding to the visual spatial points in each grid, and the conflict coding features corresponding to each radar spatial point in the grid are determined based on the vehicle driving environment and the first grid information corresponding to the radar spatial points in each grid.

[0036] In some embodiments, to reduce computing power and improve computing efficiency, for radar spatial point clouds, step 103, based on the vehicle driving environment and the first grid information corresponding to each grid, determines the conflict coding features corresponding to various types of spatial points within the grid, which can also be achieved through the following steps: For radar spatial point clouds, sectors that meet the active conditions in non-uniform coordinate sectors are identified as target sectors. The active conditions include: the sector with valid observations of visual points, front radar blocking points, and corner radar lateral points in the previous frame, and / or the sector containing the boundary spatial points of the previous frame's fused output, as well as its left and right adjacent angular sectors. Based on the vehicle driving environment and the first grid information corresponding to the radar spatial points in each grid within the target sector, the conflict coding features corresponding to the radar spatial points in the grid are determined.

[0037] In this embodiment, the front radar blocking point can be the obstacle point cloud that obstructs the drivable area as perceived by the previous forward radar sensor; the corner radar lateral point can be the effective observation point of the lateral obstacle collected by the vehicle's side corner radar sensor; and the fused output boundary space point can be the boundary point cloud of the drivable road area output after the previous radar and vision fusion processing.

[0038] In this embodiment, target sectors meeting the activity criteria are first selected from the non-uniform polar coordinate sectors where the radar spatial point cloud is located. These activity criteria include at least one of the following: the sector has valid observation data of any one of the following types in the previous frame: visual point, forward radar blocking point, or lateral radar point; the sector is the sector containing the road boundary spatial points obtained from multi-sensor fusion in the previous frame; or it is an angular sector adjacent to the left and right of the boundary sector. Then, combining the real-time driving environment of the vehicle and the first grid information of the radar spatial points stored in each grid within the target sector, the radar spatial points in each grid are analyzed to generate conflict coding features for the corresponding radar spatial points. In this way, radar point cloud calculations are performed only on sectors meeting the activity criteria, filtering out blank and invalid sectors, reducing the amount of radar point cloud processing, and lowering the real-time computing load on the vehicle; simultaneously, the fusion boundary and its neighboring sectors are covered, ensuring the integrity verification of radar data in the road traffic boundary area and reducing the problems of missed detection and missegmentation of boundary obstacles.

[0039] Step 104: Obtain the baseline log confidence of each grid. Based on the conflict coding features of various types of spatial points in each grid, the first grid information and the baseline log confidence, traverse all types of spatial points in the grid to perform multi-source data fusion and obtain the target log confidence of the grid.

[0040] In this embodiment of the application, the baseline logarithmic confidence level corresponding to each grid can be the target logarithmic confidence level corresponding to the grid in the previous frame.

[0041] In this embodiment, the target logarithmic confidence score corresponding to each grid in the previous frame is obtained and used as the baseline logarithmic confidence score for the corresponding grid in the current frame. Combining the conflict coding features of various spatial points within the current frame grid, the first grid information, and the baseline logarithmic confidence score, the multi-source sensing data fusion calculation is completed by traversing all types of spatial points in the grid, thereby obtaining the target logarithmic confidence score for each grid. Thus, the degree of contradiction in multi-sensor observations is distinguished based on conflict coding features, and differentiated fusion updates are achieved by combining grid basic information and the baseline logarithmic confidence score, fully utilizing visual and radar multi-source observation information. The use of logarithmic confidence score calculation avoids numerical overflow and improves the stability of confidence score calculation. Complete traversal of various spatial points grid by grid ensures no omission of fused information, resulting in accurate and reliable grid target confidence scores. This provides a quantitative basis for the final classification of passable areas and improves the accuracy of passable area detection under complex road conditions.

[0042] Step 105: For the same sector, traverse the target log confidence of each grid from near to far along the sector radial direction. Determine the polar radius of the first grid that meets the target log confidence condition as the target polar radius of the sector. Based on the target polar radii of all sectors, determine the current drivable area of ​​the vehicle.

[0043] In this embodiment of the application, the confidence condition includes: the target log confidence is greater than the confidence threshold, and the confidence threshold can be the log confidence threshold that has been determined by testing to have obstacles.

[0044] In this embodiment, the target logarithmic confidence scores of each grid cell are traversed radially from near to far for each sector. The polar radius corresponding to the first grid cell that meets the confidence score condition is taken as the target polar radius of that sector. The target polar radii of all sectors are aggregated, and based on the target polar radius and corresponding angle of each sector, they are converted into planar coordinates in the Cartesian coordinate system, thereby defining the boundary of the vehicle's current drivable area and generating a high-precision, highly robust drivable area. In this way, the grid confidence scores are retrieved radially from near to far along the sector, prioritizing the identification of nearby obstacle boundaries. The retrieval logic is simple and efficient, reducing the real-time computation time on the vehicle. Relying on the reliable target logarithmic confidence scores after multi-source fusion as the judgment criterion, and using the polar radius of the first grid cell that meets the standard in each sector as the boundary value, it is possible to accurately distinguish between real obstacles and false obstacles such as high-altitude areas and tunnel walls, avoiding false shrinkage of the drivable area. The target polar radii of all angle sectors are integrated to construct a complete drivable boundary, outputting a continuous and regular drivable area outline, significantly improving the accuracy and robustness of drivable area detection in complex road conditions such as tunnels and elevated roads.

[0045] In some embodiments, after obtaining the target polar radius of a sector, the target logarithmic confidence of the grid can be used as the occupancy probability of each grid, and the occupancy probability of each grid, the boundary point corresponding to the target polar radius, and the sensor source identifier, conflict coding feature, timestamp confidence and stabilization number corresponding to the boundary point can be used for data verification in subsequent frames.

[0046] In some embodiments, since the visual sensor, forward radar sensor, and corner radar sensor are all asynchronous outputs, the camera exposure time and radar scanning cycle are different. Spatial points in the same map frame may have acquisition times differing by tens or even hundreds of milliseconds. The vehicle will move forward and turn during this time difference; therefore, motion compensation is needed for each spatial point. Thus, before step 102 performs non-uniform polar coordinate sector grid projection and data structuring on the multi-source sensor data to obtain the first grid information corresponding to each spatial point, the following steps can also be performed: Step A1: Obtain time-series vehicle status information; wherein, the time-series vehicle status information includes the second acquisition time, vehicle speed, and yaw rate under different frames.

[0047] In this embodiment of the application, the time-series vehicle state information may include vehicle state information from the most recent consecutive frames. Each frame stores the same complete vehicle state tuple, including: the timestamp of the vehicle body signal output in that frame, i.e. the second acquisition time, the vehicle speed (i.e., the longitudinal driving speed of the vehicle), and the yaw rate. Of course, it may also include the yaw rate confidence level, which is used to characterize the reliability of the gyroscope / wheel speed signal.

[0048] Step A2: Starting from the second acquisition time of the spatial point cloud output by various sensors and ending with the current system time, determine the target time interval and iterate through the reference vehicle status information that falls within the target time interval in the time-series vehicle status information.

[0049] In this embodiment, the second acquisition time of the spatial point cloud output by various sensors is taken as the starting point and the current system time is taken as the ending point to determine the target time interval. The vehicle status information that falls into the target time interval in the time-series vehicle status information is traversed and determined as the reference vehicle status information.

[0050] Step A3: Based on the reference vehicle state information, determine the inter-frame difference of yaw rate between adjacent frames, and based on the inter-frame difference of yaw rate, determine the target yaw rate increment.

[0051] In this embodiment, the yaw rate difference between adjacent frames is calculated based on the yaw rate in the reference vehicle state information. Further, the target yaw rate increment is determined based on the yaw rate difference.

[0052] Here, the target yaw rate increment is determined based on the inter-frame difference in yaw rate, which can be achieved through the following process: If the inter-frame difference in yaw rate is greater than the difference threshold, the yaw rate of the current frame is limited, and the limited yaw rate of the current frame is integrated to obtain the target yaw rate increment; if the inter-frame difference in yaw rate is less than or equal to the difference threshold, the yaw rate of the current frame is integrated to obtain the target yaw rate increment. In this way, by judging the sudden changes in yaw rate through the inter-frame difference, and limiting the sudden value before integration, the jumps in yaw rate caused by sensor jitter and instantaneous interference can be filtered out, avoiding large deviations in the integration results, and preventing overall drift of spatial point boundaries caused by abnormal vehicle body signals in a single frame; direct integration under stable conditions ensures the real-time performance and accuracy of heading increment calculation, improves the stability of vehicle pose estimation, and provides reliable motion compensation data for point cloud coordinate transformation and grid matching.

[0053] Step A4: Based on the target yaw rate increment and vehicle speed, perform motion compensation on the corresponding sensor data to obtain compensated multi-source sensor data.

[0054] In this embodiment, the time difference between the first acquisition time corresponding to the spatial point cloud output by various sensors and the current system time is calculated. Within this time difference, the target yaw rate increment and vehicle speed are integrated to obtain the pose accumulation of the spatial point cloud output by that type of sensor. Based on the pose accumulation, the coordinates of the spatial point cloud output by that type of sensor are compensated to the current system time, thereby compensating the coordinates of the spatial point cloud output by all types of sensors to the current system time, thus obtaining compensated multi-source sensor data. In this way, by combining the acquisition delay of each sensor point cloud, the yaw rate increment and vehicle speed are integrated synchronously to obtain the corresponding pose accumulation. Coordinate timing compensation is performed independently for different sensor point clouds, and the multi-source point clouds are uniformly corrected to the same coordinate system at the same time in the current frame. This eliminates the point cloud spatial misalignment problem caused by the asynchronous acquisition time of each sensor, solves the grid matching deviation caused by the spatiotemporal asynchrony of multiple sensors, improves the alignment accuracy of visual and radar point cloud fusion, and provides accurate original data for spatiotemporal alignment for subsequent grid division of passable areas and conflict feature discrimination, effectively improving the stability of multi-sensor fusion perception in dynamic driving scenarios.

[0055] In some embodiments, step 103 determines the conflict coding features corresponding to various spatial points within a grid based on the vehicle driving environment and the first grid information corresponding to each grid. This can be achieved through at least one of the following methods.

[0056] Step B1: Based on the first grid information corresponding to the grid to which each visual spatial point belongs in the visual spatial point cloud and the historical frame polar radius corresponding to all sectors, determine the boundary gradient of adjacent range bands within the same sector, and the polar radius jump variable of each grid's visual spatial point in adjacent frames. If the boundary gradient is greater than the gradient threshold, the polar radius jump variable is greater than the jump variable threshold, and there is no radar spatial point corresponding to a low target within the grid, determine that the visual spatial point within the grid has the visual near-range closed-radar long-range passable feature.

[0057] In this embodiment, the historical frame polarimetric path can be the polarimetric path of the previous frame. The historical frame polarimetric path corresponding to each sector is obtained. Based on the current frame polarimetric path in the first grid information corresponding to the grid to which each visual spatial point belongs in the visual spatial point cloud, and combined with the grid attribute differences of adjacent range bands within the same sector, the boundary gradient of adjacent range bands within the same sector is determined. When it is determined that the boundary gradient of the grid exceeds a preset threshold, the polarimetric path jump variable exceeds a threshold, and there is no radar spatial point corresponding to a low target within the grid, the visual spatial point within the grid is marked as having the characteristic of visual near-range closure—radar long-range passability. Thus, by integrating spatial gradient and temporal extreme path jump dual visual features and combining them with radar low-profile target observation results for joint verification, it can accurately distinguish between false closed areas and real obstacles caused by visual imaging distortion, effectively avoiding the problem of false shrinkage of passable areas in scenarios such as backlight, uphill, downhill, shadow, road surface reflection, or ground markings; by improving the adaptability of working conditions through multi-dimensional condition joint discrimination, it also provides standardized conflict feature basis for subsequent grid confidence correction and passable boundary optimization, controlling the algorithm's computational overhead while ensuring perception and recognition accuracy, and adapting to the real-time computing needs of the vehicle terminal.

[0058] Step B2: Obtain historical frame radar spatial point clouds. Based on the historical frame radar spatial point clouds and the radar spatial point clouds, statistically analyze the radar grid information of each grid. For each grid, if the radar grid information satisfies both the first and second conditions, it is determined that the radar spatial points within the grid have consistent characteristics. If the radar grid information satisfies the first condition but not the second condition, it is determined that the radar spatial points within the grid have historical jump characteristics. The radar grid information includes the number of times the radar boundary points appear when determining the drivable area corresponding to the historical frame, the polar radius variance, and the directional stability. The first condition includes: the number of occurrences is greater than a threshold value. The second condition includes: the polar radius variance is greater than a threshold value, and the directional stability is greater than a stability threshold value.

[0059] In this embodiment of the application, the historical frame radar spatial point cloud can be the radar spatial point cloud collected and stored from multiple consecutive frames before the current frame, such as 30 frames, to provide a basis for temporal dimension observation.

[0060] In this embodiment, the radar grid information can be a grid time-series quantization index obtained based on the statistical analysis of radar point clouds in historical frames and the current frame, including but not limited to the number of radar boundary point occurrences, polar radius variance, and directional stability. Here, the number of radar boundary point occurrences can be the total number of times radar boundary observation points within the grid have appeared during the construction of the drivable area in historical frames, reflecting the long-term existence of obstacles; directional stability is used to characterize the consistency of the angle distribution of radar points across multiple frames, with a higher value indicating smaller changes in the obstacle observation angle.

[0061] In this embodiment, historical frame radar spatial point clouds are obtained. Combining the historical frame radar spatial point clouds with the current frame radar spatial point clouds, radar grid information corresponding to each grid is statistically analyzed. For each grid, feature determination logic is executed: if the number of radar boundary point occurrences is greater than a preset threshold, and the polar radius variance is greater than a variance threshold and the directional stability is greater than a stability threshold, the grid information satisfies both the first and second conditions, and the radar spatial points within the grid are determined to have consistency characteristics. If the number of radar boundary point occurrences is greater than a preset threshold, but the polar radius variance is less than or equal to the variance threshold and the directional stability is less than or equal to the stability threshold, the grid information satisfies only the first condition and not the second condition, and the radar spatial points within the grid are determined to have historical jump characteristics. Thus, by fusing the current frame and multiple frames of historical radar point clouds, multi-dimensional radar grid information including the number of boundary point occurrences, polar radius variance, and directional stability is generated. Two levels of determination conditions are used to distinguish radar grids with consistency characteristics from those with historical jump characteristics.

[0062] Step B3: Obtain the polar radius difference between visual spatial points and radar spatial points located in the same grid. If the polar radius difference is greater than the polar radius difference threshold, determine that the radar spatial points in the grid have radar near-range blocking - visual long-range passability characteristics; if the polar radius difference is less than the polar radius difference threshold, determine that the visual spatial points in the sector have visual near-range closure - radar long-range passability characteristics.

[0063] In this embodiment, by calculating the polar radius difference between visual spatial points and radar spatial points within the same grid, a quantitative comparison of the radial distance between the two heterogeneous sensors is constructed. Using the polar radius difference threshold as the judgment boundary, two typical perception difference scenarios are accurately distinguished: when the polar radius difference is greater than the threshold, it indicates that the radar detects a near-range obstruction target while the visual perception still perceives a passable space at a greater distance. This determines that the radar spatial points within the grid have the characteristics of near-range radar obstruction and far-range visual passability, thereby effectively utilizing the radar's high sensitivity to near-range hard obstacles to capture potential risks. When the polar radius difference is less than or equal to the threshold, it indicates that the visual point cloud has perceived a closed boundary at close range while the radar still determines it as an open area at a greater distance. This determines that the visual spatial points within the grid have the characteristics of near-range visual obstruction and far-range radar passability, fully leveraging the advantage of vision in the fine perception of texture boundaries. In this way, by using the difference in polar radius as a unified metric, the outputs of heterogeneous sensors are mapped to the same radial coordinate system for difference quantification. This avoids the reliance on absolute confidence in the results of a single sensor and provides clear perceptual semantics for downstream decision-making through symbolic feature annotation. At the same time, the overall operation only involves subtraction and threshold comparison, which is computationally lightweight, logically clear, and easy to deploy in real time on embedded platforms. This significantly improves the interpretability and engineering practicality of multi-sensor fusion perception in complex traffic scenarios.

[0064] Step B4: When the vehicle is traveling in a tunnel environment, the tunnel area is determined based on high-precision map information and tunnel marker information output by visual sensors, and radar spatial points within the tunnel area are determined to have suspected tunnel wall features.

[0065] In this embodiment of the application, the tunnel signage information may be visual features such as tunnel entrances and tunnel sidewalls identified by a visual sensor.

[0066] In this embodiment, the drivable area determination device pre-stores high-precision map information. When the vehicle is traveling in a tunnel environment, the tunnel area is delineated by fusing high-precision map data with tunnel marker information output by visual sensors. Radar spatial points falling within this tunnel area are uniformly marked as suspected tunnel wall features. Thus, this embodiment accurately delineates the tunnel range by combining high-precision map positioning and visual tunnel marker information, quickly assigning suspected tunnel wall features to radar points within the tunnel. This allows for early differentiation between radar virtual points generated by tunnel wall reflections and real road obstacles, reducing miscompression of drivable areas in tunnel scenarios. Relying on map and visual fusion to locate tunnel sections results in higher positioning reliability and provides a tunnel-specific correction basis for subsequent grid conflict coding and confidence attenuation, significantly improving the detection accuracy of drivable areas in low-light and multi-reflection tunnel conditions.

[0067] Step B5: Based on the forward-looking high-altitude marker information output by the visual sensor and the radar high-altitude marker position output by the radar sensor, determine the high-altitude region where the high-altitude target is located, and determine that the radar spatial points located within the high-altitude region have suspected high-altitude target characteristics.

[0068] In this embodiment, by fusing forward-looking high-altitude marker information output from a visual sensor with high-altitude marker positions output from a radar sensor, the high-altitude region corresponding to the high-altitude target is delineated, and radar spatial points within this high-altitude region are marked as suspected features of the high-altitude target. Thus, by collaboratively delineating high-altitude regions using both visual and radar information, radar reflection points corresponding to high-altitude objects such as bridges, road signs, and traffic lights can be quickly distinguished. This provides dedicated feature identification for high-altitude clutter filtering, avoiding misjudging non-obstacle objects as road obstructions, effectively expanding the detection range of passable areas, and improving the accuracy of perception results in scenarios with multiple high-altitude facilities such as elevated roads and urban roads.

[0069] Step B6: Obtain the current frame system time. If the time difference between the current frame system time and the first acquisition time is greater than the time difference threshold, determine that various spatial points within the grid have time expiration characteristics.

[0070] In this embodiment, the current frame system time is obtained, and the difference between the current frame system time and the first acquisition time of each spatial point in each grid is calculated. If the time difference exceeds a preset time difference threshold, the spatial point in the grid is determined to have time expiration characteristics. In this way, by using time threshold verification to distinguish failed historical observation points, spatial points that have not been updated for a long time and have lost their reference value are promptly removed. This avoids expired data from continuously interfering with the calculation of grid confidence and the determination of passable areas, reduces the problems of false detection and missegmentation caused by old observations, and ensures the real-time effectiveness of multi-sensor fusion sensing results.

[0071] In some embodiments, step 104 involves traversing all types of spatial points within a grid to perform multi-source data fusion based on the conflict coding features corresponding to various types of spatial points in each grid, the first grid information, and the baseline log confidence, to obtain the target log confidence of the grid. This can be achieved through the following steps.

[0072] Step C1: Based on the current system time and the first acquisition time, determine the reliability of the timestamps corresponding to various spatial points.

[0073] In this embodiment of the application, the timestamp reliability can be obtained based on the reliability parameter obtained from the acquisition time of the spatial point cloud and the time difference of the current frame. Spatial points whose time has expired correspond to lower reliability.

[0074] In this embodiment of the application, based on the current system time and the first acquisition time, the reliability of the timestamps corresponding to various spatial points is determined by the first formula corresponding to various spatial points.

[0075] Here, the first calculation formula can be expressed by the following formula (1).

[0076]

[0077] in, For timestamp reliability, , The system time for the current frame. The sampling time of the sensor corresponding to each spatial point is the first acquisition time. For the time decay constants corresponding to various sensors, it should be noted that the configurations for visual sensors, forward-facing radar sensors, and corner radar sensors are different. This is to match their respective sampling delay characteristics. It should be noted that... The closer it gets to 0, The closer it is to 1, the higher the reliability of the characterization observation.

[0078] Step C2: Based on timestamp reliability, conflict coding features, and first grid information, determine the gating coefficients corresponding to each spatial point within the grid.

[0079] In this embodiment, the gating coefficient is used to dynamically adjust the contribution of each spatial point to the grid confidence during multi-source data fusion.

[0080] In this embodiment, the gating coefficients corresponding to each spatial point within the grid are determined based on timestamp reliability, conflict coding features, and first grid information. This can be achieved through the following process: Based on the vehicle driving environment, the sensor basic reliability factors for various spatial points within the grid are determined; based on sector identifiers and distance band identifiers, the spatial distance factors for various spatial points within the grid are determined; the conflict coding feature factors corresponding to the conflict coding features and the historical stability factors corresponding to the historical stability are obtained; based on timestamp reliability, sensor basic reliability factors, spatial distance factors, conflict coding feature factors, and historical stability factors, the gating coefficients corresponding to each spatial point within the grid are determined.

[0081] In this embodiment of the application, the sensor basic reliability factor is used to characterize the observation reliability benchmark of the corresponding sensor under the working condition. The sensor basic reliability factor can be determined based on the sensor basic reliability in the first grid information corresponding to the visual spatial point. For example, the sensor basic reliability factor can be the sensor basic reliability, or the sensor basic reliability can be weighted to determine the sensor basic reliability factor.

[0082] In this embodiment, the spatial distance factor is used to reflect the attenuation characteristics of the detection accuracy of the radar / visual sensor as the detection distance and angle change. The spatial distance factor can be a weighting coefficient calculated from the sector and range band identifiers.

[0083] In this embodiment, the conflict coding feature factor is used to suppress the influence of contradictory observation data. The conflict coding feature factor can be a weight coefficient that corresponds one-to-one with the conflict coding of multi-source observations within the raster. The higher the degree of conflict, the lower the value of this factor.

[0084] In this embodiment, the historical stability factor is used to generate weights based on the temporal characteristics of radar spatial point clouds or historical stability counts. Long-term stable observation points are given higher factors, while observation points with temporal fluctuations have lower factor weights.

[0085] In this embodiment, the gating coefficient can be a single-point weighted coefficient calculated by integrating multiple dimensions of factors such as timeliness, sensor operating conditions, spatial location, observation conflicts, and temporal stability. Specifically, the gating coefficient of a spatial point is determined based on the timestamp reliability, sensor basic reliability factor, spatial distance factor, conflict coding characteristic factor, and historical stability factor of each spatial point. It should be noted that the gating coefficient of a spatial point can be obtained by summing the timestamp reliability, sensor basic reliability factor, spatial distance factor, conflict coding characteristic factor, and historical stability factor of each spatial point, or by multiplying the timestamp reliability, sensor basic reliability factor, spatial distance factor, conflict coding characteristic factor, and historical stability factor of each spatial point. Of course, it can also be obtained by summing at least a portion of the timestamp reliability, sensor basic reliability factor, spatial distance factor, conflict coding characteristic factor, and historical stability factor of each spatial point, multiplying the remaining portions, and then adding the two. This application does not impose specific limitations on this.

[0086] In this embodiment, the basic reliability factor of each sensor spatial point is determined based on the vehicle driving environment; the spatial distance factor is calculated based on the sector identifier and distance band identifier; and the conflict coding feature factor and historical stability factor are extracted simultaneously. The gating coefficient corresponding to each spatial point within the grid is solved by comprehensively considering the timestamp reliability, sensor basic reliability factor, spatial distance factor, conflict coding feature factor, and historical stability factor. Thus, by setting weighting factors in multiple dimensions, the reliability of a single point is quantified comprehensively from the perspectives of scene conditions, detection distance, observation conflict, temporal stability, and data timeliness, and the gating coefficient is generated in a refined manner. This enables adaptive weighted fusion of multi-source point clouds under different scenes, distances, and observation states, weakening the interference of low-reliability observations on grid confidence, strengthening the weight of effective observations, significantly improving the grid confidence update accuracy in complex tunnel, backlight, and long-distance scenarios, and optimizing the robustness of passable area detection.

[0087] Step C3: Obtain the logarithmic confidence increments corresponding to various spatial points within the raster.

[0088] In this embodiment, the log-odds increment is the base Bayesian logarithm increment output by the inversion model corresponding to different types of sensors. It should be noted that, based on the hardware detection physical characteristics of each type of sensor, a corresponding independent physical inversion model is set up. Specifically, these are the visual inversion model for visual sensors, the forward radar inversion model for forward radar sensors, and the angular radar inversion model for angular radar sensors. Each model is customized according to its own hardware detection physical characteristics, and outputs a base log-odds increment adapted to that sensor, which serves as the raw input for subsequent gated weighting and grid confidence updates.

[0089] In this embodiment, the input information of the inversion model corresponding to different types of sensors is different, but the output information is the same. That is, the output information of the inversion model corresponding to different types of sensors includes not only the log confidence increment corresponding to the spatial point cloud, but also the boundary direction and distance relationship.

[0090] The logarithmic confidence increment can be the logarithmic increment output by the inversion model based on the input information, determining whether a spatial point is passable and open or blocked by obstacles. The fixed increments include: strongly passable, weakly passable, slightly open, neutral, slightly blocked, strongly blocked, and extremely strongly blocked. Here, if radar is blocked, a positive increment is output to increase the grid occupancy confidence; if visually passable, a negative increment is output to decrease the grid occupancy confidence. It should be noted that the magnitude of the logarithmic confidence increment differs for visual and radar measurements at the same observation intensity due to differences in ranging confidence.

[0091] Among them, the boundary direction represents the radial direction of the current observed polar radius in the sector, and is used for subsequent radial monotonic constraint post-processing to distinguish between two directions: the near-distance inward contraction boundary and the far-distance widening of the passable area.

[0092] In one feasible approach, a visual inversion model is generated based on the camera's optical imaging characteristics. The input information to the visual inversion model includes, but is not limited to: image illumination intensity, shadow area markings, slope markings, road surface texture clarity, and image blur. The output information includes the basic logarithmic confidence score for each visual spatial point, where the basic logarithmic confidence score for each visual spatial point is a negative increment. When the road surface texture is clear and the illumination is normal, the absolute value of the negative increment is large, fully acknowledging the smooth observation; under shadow / backlight conditions, the amplitude of the negative increment is automatically reduced to weaken the impact of visual false closure.

[0093] In one feasible approach, a forward radar inversion model is generated based on the long-range ranging and electromagnetic wave reflection characteristics of the forward radar. The input information of the forward radar inversion model includes, but is not limited to: echo elevation height determined based on the forward radar spatial point cloud, long-range ranging variance, continuous equidistant wall reflection characteristics, and high-altitude target echo markers. The output information includes: the logarithmic confidence increment and boundary direction corresponding to each forward radar spatial point, wherein the logarithmic confidence increment corresponding to the forward radar spatial point is a positive increment. When there are low-to-medium altitude ground obstacles directly in front, the positive increment amplitude is large; for long-range, high-altitude, and wall reflection points, the positive increment amplitude automatically decreases, and the boundary direction is used to correct the longitudinal boundaries of the vehicle.

[0094] In one feasible approach, an angle radar inversion model is generated based on the hardware characteristics of the angle radar's lateral short-range detection. The input information of the angle radar inversion model includes: lateral guardrail continuous reflection markers determined based on the angle radar spatial point cloud, the markings of the overlapping area with the front radar angle, and the lateral clutter intensity. The output information includes: the logarithmic confidence increment and boundary direction corresponding to each angle radar spatial point, wherein the logarithmic confidence increment corresponding to each angle radar spatial point is a small positive increment; in the forward overlapping sector with the front radar, the basic increment is forcibly suppressed and only serves as an auxiliary point, and cannot independently dominate the forward boundary; the increment amplitude in the lateral region is normal; the boundary direction is used to correct the left and right lateral boundaries of the vehicle and does not interfere with the forward longitudinal distance.

[0095] In this embodiment, based on the inversion models corresponding to each type of sensor, the logarithmic confidence increments for each type of spatial point within each grid are obtained. Combining the conflict coding features corresponding to each type of spatial point within the grid, the first grid information, and the logarithmic confidence increments, the multi-source sensing data fusion calculation is completed by traversing all types of spatial points in the grid, thereby obtaining the target logarithmic confidence for each grid. Thus, the degree of contradiction in multi-sensor observations is distinguished based on conflict coding features, and differentiated fusion updates are achieved by combining grid basic information and logarithmic confidence increments, fully utilizing visual and radar multi-source observation information. The use of logarithmic confidence calculation avoids numerical overflow and improves the stability of confidence calculation. Complete traversal of each type of spatial point ensures no omission of fused information, resulting in accurate and reliable grid target confidence, providing a quantitative basis for the final classification of passable areas and improving the accuracy of passable area detection under complex road conditions.

[0096] Step C4: Based on the logarithmic confidence increment and gating coefficient corresponding to various types of spatial points within the raster, the baseline logarithmic confidence is updated in a fixed-point manner until all types of spatial points within the raster are traversed to achieve multi-source data fusion, thereby obtaining the target logarithmic confidence of the raster.

[0097] In this embodiment, each grid includes at least one type of spatial point. Based on the logarithmic confidence increment and gating coefficient corresponding to the first type of spatial point in the grid, the baseline logarithmic confidence is updated by the fixed-point representation of that type of spatial point to obtain an intermediate logarithmic confidence of the grid. Further, based on the logarithmic confidence increment and gating coefficient corresponding to the second type of spatial point in the grid, the intermediate logarithmic confidence is updated by the fixed-point representation of that type of spatial point to obtain another intermediate logarithmic confidence of the grid, until all types of spatial points in the grid have been traversed, thereby realizing multi-source data fusion and obtaining the target logarithmic confidence of the grid.

[0098] In some embodiments, step C4 updates the baseline logarithmic confidence level based on the logarithmic confidence increment and gating coefficient corresponding to various types of spatial points within the grid, until all types of spatial points within the grid are traversed to achieve multi-source data fusion, thereby obtaining the target logarithmic confidence level of the grid. This can be achieved through the following process: If the grid includes visual spatial points and radar spatial points, the baseline logarithmic confidence level of the grid is updated based on the logarithmic confidence increment and gating coefficient corresponding to the visual spatial points within the grid to obtain the first intermediate logarithmic confidence level of the grid; the first intermediate logarithmic confidence level of the grid is updated based on the logarithmic confidence increment and gating coefficient corresponding to the radar spatial points within the grid to obtain the target logarithmic confidence level.

[0099] In this embodiment, the various spatial points within the grid include visual spatial points and radar spatial points. Based on the logarithmic confidence increment and gating coefficient corresponding to the visual spatial points within the grid, the baseline logarithmic confidence of the grid is updated using the following formula (2) to obtain the intermediate logarithmic confidence. Based on the logarithmic confidence increment and gating coefficient corresponding to the radar spatial points within the grid, the intermediate logarithmic confidence of the grid is updated using the following formula (2) to obtain the target logarithmic confidence of the grid.

[0100] Wherein, formula (2) can be

[0101] Where L is the log-confidence level of the raster. The gating coefficient corresponding to the spatial point. This represents the logarithmic confidence increment corresponding to the spatial point.

[0102] As described above, the embodiments of this application sequentially update the confidence levels of visual and radar point clouds step by step, dynamically adjusting the confidence level correction magnitude of a single type of observation based on their respective matching gating coefficients; first, visual observations are fused and then radar observations are superimposed, achieving orderly weighted fusion of the two types of sensor data, which can adaptively reduce the interference of abnormal observations on the grid confidence level according to the confidence level of the point cloud; the hierarchical update logic is clear and the calculation process is modular, which can accurately correct the grid baseline confidence level and output the target logarithmic confidence level that fits the real road conditions, effectively improving the accuracy of the traversable area division and the perception stability of complex scenes.

[0103] In some embodiments, if the grid includes visual spatial points and radar spatial points, the baseline logarithmic confidence of the grid is updated based on the logarithmic confidence increment and gating coefficient corresponding to the visual spatial points within the grid to obtain the first intermediate logarithmic confidence of the grid; the target logarithmic confidence is updated based on the logarithmic confidence increment and gating coefficient corresponding to the radar spatial points within the grid. This can be achieved through the following process: If the grid includes visual spatial points and radar spatial points, based on the logarithmic confidence increment and gating coefficient corresponding to the visual spatial points within the grid, the baseline logarithmic confidence of the grid is updated based on the logarithmic confidence increment and gating coefficient corresponding to the visual spatial points within the grid to obtain the target logarithmic confidence. The baseline logarithmic confidence of the grid is updated using confidence increments and gating coefficients to obtain the second intermediate logarithmic confidence of the grid. Based on the second intermediate logarithmic confidence and / or the conflict coding features corresponding to the visual spatial point, the first intermediate logarithmic confidence is determined. Based on the logarithmic confidence increment and gating coefficients corresponding to the radar spatial point within the grid, the first intermediate logarithmic confidence of the grid is updated to obtain the third intermediate logarithmic confidence. Based on the third intermediate logarithmic confidence and / or the conflict coding features corresponding to the radar spatial point, the target logarithmic confidence is determined.

[0104] In this embodiment, the grid includes visual spatial points and radar spatial points. Based on the logarithmic confidence increment and gating coefficient corresponding to the visual spatial points in the grid, the baseline logarithmic confidence of the grid is updated using the above formula (2) to obtain the second intermediate logarithmic confidence of the grid. If the conflict coding feature corresponding to the visual spatial point is a suspected feature of a high-altitude target or a suspected feature of a tunnel wall, the preset minimum logarithmic confidence is directly determined as the first intermediate logarithmic confidence of the grid. If the conflict coding feature corresponding to the visual spatial point is not a suspected feature of a high-altitude target or a suspected feature of a tunnel wall, the first intermediate logarithmic confidence of the grid is determined based on the second intermediate logarithmic confidence using the following formula (3). Similarly, based on the logarithmic confidence increment and gating coefficient corresponding to the radar spatial point in the grid, the first intermediate logarithmic confidence of the grid is updated using the above formula (2) to obtain the fourth intermediate logarithmic confidence. If the collision coding feature corresponding to the radar spatial point is a suspected feature of high-altitude target or a suspected feature of tunnel wall, the pre-set minimum logarithmic confidence is directly determined as the third intermediate logarithmic confidence of the grid. If the collision coding feature corresponding to the radar spatial point is not a suspected feature of high-altitude target or a suspected feature of tunnel wall, the target logarithmic confidence of the grid is determined based on the third intermediate logarithmic confidence using the following formula (3).

[0105] Formula (3) can be:

[0106] Where L is the logarithmic confidence level. For a pre-set minimum log-confidence level, The maximum log confidence level is set in advance. This is the trimming function, which means: if L on the right side of the equals sign satisfies... Then Assign the value to L on the left side of the equals sign; if L on the right side of the equals sign satisfies Then Assign the value to L on the left side of the equals sign; if L on the right side of the equals sign satisfies If so, then the value of L on the right side of the equals sign will be assigned to L on the left side of the equals sign.

[0107] As described above, this application's embodiments employ dedicated confidence degradation logic for invalid obstacle points such as tunnel walls and high-altitude markers. Once the corresponding feature is identified, the minimum logarithmic confidence level is directly assigned, quickly reducing the interference of virtual obstacle points on the grid confidence level and preventing the erroneous compression of passable areas in tunnels and elevated road sections. Visual and radar data independently verify conflict features and iterate confidence levels in stages. Abnormal observations from the two types of sensors can be suppressed separately without mutual superposition and amplification of errors, resulting in stronger multi-source fusion fault tolerance.

[0108] The process of the method provided in this application will be illustrated below through a specific embodiment.

[0109] This application provides a multi-sensor freespace point fusion method based on a low-computing-power platform (corresponding to the aforementioned drivable area determination method). This method first constructs a non-uniform sparse polar coordinate grid aligned with the radar angular resolution. Second, it converts the freespace points from the visual, front radar, and corner radar into point cells with timestamp confidence and source masks. Then, it generates conflict types based on scene and spatial consistency. Finally, it uses gated weighted Bayesian log-odds updates, radial monotonic constraints, and historical boundary backfeeding to output the fusion result. The method includes the following steps: Step S1: Data preprocessing.

[0110] Here, visual sensor data and radar sensor data are obtained.

[0111] Here, for visual sensor data, the point cloud of the visually passable area (Freespace) output by the forward-looking camera is received. Visual points retain their original Cartesian coordinates (X, Y). A mapping table is pre-established between Cartesian coordinates and polar coordinates at different angular resolutions for different types of radar sensors. Using this mapping table, the sector number and range band number corresponding to the coordinates of each visual point are looked up. Then, a structure Ev={r,theta,src,t,mask,q} is generated for the visual Freespace points, thus achieving preprocessing of the visual sensor data. Here, r is the polar radius, theta is the sector number, src represents the visual source, t represents the acquisition time, mask represents the target type / dynamic attribute / occlusion attribute, and q represents the visual baseline quality score.

[0112] Here, radar sensor data is used, including forward millimeter-wave radar sensor data and corner radar sensor data. The forward millimeter-wave radar sensor data consists of 240 forward radar freespace points arranged at 0.5° intervals, generated from data collection of the forward region directly in front of the vehicle by the forward millimeter-wave radar sensor. During preprocessing, the 0.5-degree angular resolution of the forward millimeter-wave radar sensor is used as the fine sector width of the forward region, and the nearest boundary, farthest passable distance, echo height marker, and historical stability count are recorded for each sector. The corner radar sensor data consists of 120 corner radar freespace points arranged at 1° intervals, generated from data collection of the left and right lateral regions of the vehicle by the corner radar sensor. During preprocessing, the corner radar freespace points are mapped to coarse lateral sectors, and a one-to-many mapping table is established within the angular range overlapping with the front radar, allowing the lateral points of the corner radar to reinforce the visual side edges but not directly cover the stable boundary directly in front of the front radar. Furthermore, the radar sensor data is converted into a unified freespace point structure.

[0113] It should be noted that radar FreeSpace points are no longer uniformly converted to Cartesian global grids; instead, the radar's original polar coordinate representation is retained first. For local grids that need to be compared with visual points, the polar coordinate to Cartesian coordinate conversion is only performed within the active sector, thereby reducing the number of trigonometric function calls on low-computing-power platforms.

[0114] As can be seen from the above, the incoming forward-looking, forward radar, and corner radar sensor data are converted into a unified freespace point structure and projected onto a non-uniform polar coordinate grid according to the sensor's native angular resolution; active indexes are only established for sectors that have been observed or affected by historical boundaries, avoiding full traversal of the two-dimensional grid.

[0115] Step S2: Data verification and conflict coding feature generation.

[0116] The preprocessed sensor data is verified based on scene and physical characteristics. Unlike simple filtering, this embodiment generates a conflict code C for each suspicious observation. The conflict code includes at least C0 consistency feature, C1 visual near-range closure / radar long-range passage, C2 radar near-range obstruction / visual long-range passage, C3 suspected tunnel wall, C4 suspected high-altitude target, C5 time expiration, and C6 historical jump.

[0117] Specifically, for visual data verification, verification is performed by combining the target type, dynamic information, and road plane continuity of the visual point cloud. For false detections of visual road closures caused by uphill, downhill, shadows, or ground markings, the boundary gradient and continuous frame jump variables of adjacent distance bands within the same sector are calculated. When the visual boundary changes abruptly from far to near in a short period of time and the radar has no corresponding blocking point, the visual point is marked as C1 instead of being directly used as the final boundary.

[0118] Specifically, for radar data verification, the saved historical frames of radar data are traversed to calculate the occurrence frequency, polar radius variance, and directional stability of the radar boundary within the current sector. Only radar FreeSpace points that simultaneously meet the occurrence frequency threshold and variance threshold are marked as having strong consistency features; if only the occurrence frequency threshold is met but the variance is too large, it is downgraded to the C6 historical jump feature.

[0119] Specifically, for tunnel scenarios, the system combines visual tunnel markers, high-precision map information, and the symmetrical reflection characteristics of the left and right side radars to identify suspected tunnel wall areas. When the radar polarimeters of multiple adjacent sectors form a continuous wall shape, the visual freespace boundary is inconsistent with it, and the vehicle's yaw rate is small, the area is marked as C3, and its effect on shrinking the passable boundary is restricted.

[0120] Specifically, for high-altitude false detection filtering, and to address the false detections of traffic signs, gantry cranes, and other high-altitude objects by millimeter-wave radar, the system verifies the detection by combining the sign frame identified by the forward-looking system, the radar's high-altitude marker position, and the elevation compensation distance of the current grid. If the radar point is within the projection range of the high-altitude target, it is marked as C4, and its log-odds increment is set to zero or attenuated to a preset lower limit.

[0121] As can be seen from the above, sensor data is verified based on physical characteristics, spatial location, historical stability, and special scenarios, and conflict codes such as visually passable, radar blocked, high-altitude suspected, tunnel wall suspected, time expired, and near-far conflict are generated for each grid, instead of simply retaining or filtering.

[0122] Step S3: Motion compensation and time synchronization.

[0123] Because the input times of each sensor are not perfectly aligned with the system fusion time, motion compensation is required for the freespace points of each sensor. The compensation target is not the entire map, but the freespace points within the active polar coordinate sectors, thus meeting the real-time requirements of low-computing-power platforms.

[0124] First, a circular array is used to store the vehicle information of the most recent 30 frames. The vehicle information includes timestamp, vehicle speed, yaw rate, and yaw rate confidence. When the yaw rate of adjacent frames changes abruptly, the yaw rate increment after amplitude limiting is used for compensation to avoid the overall drift of the freespace boundary caused by abnormal vehicle signal in a single frame. Further, the cumulative pose of freespace points between sensor time and system time is calculated using the vehicle information, and the positions of freespace points of all sensors are compensated to the current system time. At the same time, the timestamp confidence lambda_t is calculated using the above formula (1). lambda_t decays as the time difference increases and participates in the subsequent Bayesian update weights.

[0125] As can be seen from the above, the compensation position of each freespace point is calculated using the timestamp of each sensor and the system time, the vehicle speed, yaw rate, and yaw rate increment after low-pass filtering; at the same time, a timestamp reliability factor is generated based on the time difference, so that observations with large asynchronousities participate in the fusion but the update intensity is reduced.

[0126] Step S4: Bayesian log-odds update for conflict point gating.

[0127] First, inversion models conforming to the physical detection characteristics of vision, front radar, and corner radar are pre-generated. The basic log-odds increment DeltaL_s(e) for each freespace point of that sensor type is output using the inversion models corresponding to each sensor type. Second, since each active polar coordinate grid includes time-synchronized multi-sensor FreeSpace points, the log-odds confidence L of the previous frame for each active polar coordinate grid is obtained and sequentially passed to the FreeSpace points of each sensor type within that active polar coordinate grid. For each sensor type's FreeSpace point within the active polar coordinate grid, based on the current collision type C, timestamp confidence lambda_t, polar radius r, sector number theta, and historical stability h corresponding to that FreeSpace point, the gating coefficient g is calculated and finally calculated according to L=clip(L+g). DeltaL_s,Lmin,Lmax) performs fixed-point updates on the log-odds confidence L of the raster. It should be noted that when C is a high-altitude suspected case or a tunnel wall suspected case, g is restricted to the low-value interval Lmin.

[0128] Here, the gating coefficients corresponding to the FreeSpace points can be obtained based on sensor characteristics, timestamp reliability, spatial location, collision type, and historical stability. Specifically, the factors corresponding to the sensor characteristics, spatial location, collision type, and historical stability of the FreeSpace points are obtained, and the multiple factors are multiplied by the timestamp reliability to obtain the gating coefficients corresponding to the FreeSpace points.

[0129] Here, regarding sensor characteristic factors, vision has high lateral boundary accuracy under normal lighting and stable road texture, but it is prone to false detection of close-range closures under shadows, slopes, and strong backlighting; front radar is relatively stable at long distances directly in front, but is sensitive to high-altitude objects and tunnel walls; corner radar is reliable for lateral boundaries, but only serves as an auxiliary freespace point in the forward overlap area.

[0130] Here, for spatial location factors, the confidence level is adjusted based on the angle, range band, and overlap of the detection point in the sensor coordinate system. For forward fine sectors, the consistency features between forward radar and vision are prioritized, while for lateral sectors, the consistency features between corner radar and visual edges are prioritized. Boundaries with excessively large jumps across sectors need to be confirmed through radial monotonic constraints and smoothing.

[0131] Here, regarding environmental context factors, in tunnel scenarios, instead of directly reducing or increasing the overall weight of a particular sensor, we identify continuous wall morphology, left-right reflection symmetry, and visual boundary consistency. When a radar point matches the suspected tunnel wall feature C3, we only reduce the update increment corresponding to that conflict code, retaining it as a weak freespace point of an unconfirmed obstacle.

[0132] Here, for high-altitude false detection filtering, when the context information indicates that there is a sign, gantry, or height restriction pole ahead, and the radar point has a high-altitude mark or coincides with the visual high-altitude frame projection, the point is written into the history as a suspected high-altitude target C4 feature but does not participate in boundary shrinkage; if a low-altitude target freespace point appears in subsequent consecutive frames, the suppression is lifted again.

[0133] Finally, the final log-odds confidence L of each active polar coordinate grid is used to determine the final occupancy probability of each grid. Here, after fusing the time-synchronized multi-sensor FreeSpace points within each active polar coordinate grid, each grid maintains a state S={L,r_f,m_src,C,h}, where L is the occupancy probability log-odds, r_f is the fused freespace boundary polar radius, m_src is the sensor source identifier, C is the current collision coding type, and h is the historical stability. It should be noted that continuous and stable near-boundary points are prioritized in the output, while suppressed anomalous FreeSpace points are retained for diagnosis and verification in the next frame.

[0134] As described above, the freespace of the multi-sensor system after time synchronization is distributed within each polar coordinate grid. For each active grid, a gating weight is calculated based on the sensor source, collision type, spatial location, timestamp reliability, and scene state. The occupancy probability is then updated using a fixed-point log-odds format to avoid the computational overhead caused by repeated normalization of floating-point probabilities.

[0135] Step S5: Historical information saving, radial constraints and output.

[0136] Here, the radar and visual FreeSpace points of the current frame, after verification and gating, are cyclically saved to a history array. The history array not only stores the polar radius but also the collision code, source mask, timestamp reliability, and stability count, used for data verification, tunnel wall identification, and high-altitude false detection removal in subsequent frames. Then, radial monotonic constraints and short-window smoothing are applied to the fused boundary. Specifically, strong blocking points at close range within the same sector can suppress passable points at long distances, and boundary jumps between adjacent sectors exceeding a threshold require confirmation by historical stability. The final output is a high-precision, highly robust passable area (FreeSpace) result.

[0137] As can be seen from the above, the radar and visual freespace points of the current frame after gating are saved, and the current output is subjected to radial monotonic constraints and short-term backfeeding in combination with the historical boundary stability. The final output includes passable boundary points, grid confidence, sensor source and abnormal conflict flags.

[0138] This application provides a device for determining a drivable area, referring to... Figure 3 As shown, Figure 3 This is a schematic diagram of a drivable area determination device provided in an embodiment of this application. The drivable area determination device 3 includes: The acquisition module 301 is used to acquire multi-source sensor data around the vehicle and the vehicle driving environment; wherein, the multi-source sensor data includes: visual spatial point cloud and / or target attribute information output by the visual sensor, and radar spatial point cloud output by the radar sensor, and the spatial points are the boundary points of the drivable area identified by the sensors. Processing module 302 is used to perform non-uniform polar coordinate sector grid projection and data structuring processing on multi-source sensor data to obtain the first grid information corresponding to the same grid to which various spatial points belong; wherein, the first grid information includes at least one of the following: the polar radius of various spatial points under the non-uniform polar coordinate grid, sector identifier, distance band identifier under the sector corresponding to the sector identifier, sensor source identifier, first acquisition time, target attribute information, sensor basic reliability and historical stability count; The determination module 303 is used to determine the conflict coding features corresponding to various spatial points within the grid based on the vehicle driving environment and the first grid information corresponding to each grid. The module 301 is also used to obtain the baseline log confidence level corresponding to each grid cell; The processing module 302 is also used to traverse all types of spatial points in the grid to perform multi-source data fusion based on the conflict coding features, first grid information and baseline log confidence of each type of spatial point in each grid, and to obtain the target log confidence of the grid. The determination module 303 is also used to traverse the target log confidence of each grid along the radial direction of the same sector from near to far, determine the polar radius of the first grid that meets the confidence condition as the target polar radius of the sector, and determine the current drivable area of ​​the vehicle based on the target polar radii of all sectors.

[0139] This application provides a hardware entity diagram of a drivable area determination device, such as... Figure 4 As shown, the hardware entity of the drivable area determination device 4 includes: a processor 401 and a memory 402, wherein the memory 402 stores a computer program that can run on the processor 401, and when the processor 401 executes the computer program, it implements some or all of the steps in the drivable area determination method as described in the above embodiments.

[0140] The memory 402 stores computer programs that can run on the processor. The memory 402 is configured to store instructions and applications that can be executed by the processor 401. It can also cache data to be processed or already processed by the various modules in the processor 401 and the drivable area determination device 4 (e.g., image data, audio data, voice communication data and video communication data). It can be implemented by flash memory or random access memory (RAM).

[0141] Specifically, when the processor 401 executes the program, it implements the steps of the drivable area determination execution method described above. The processor 401 typically controls the overall operation of the drivable area determination device 4.

[0142] This application provides a vehicle, referring to... Figure 5 As shown, Figure 5 This is a schematic diagram of the structure of a vehicle provided in an embodiment of this application. The vehicle 5 includes the aforementioned drivable area determination device 4.

[0143] This application provides a computer-readable storage medium storing one or more computer programs, which can be executed by one or more processors to implement some or all of the steps in the above-described method. The storage medium can be transient or non-transient.

[0144] This application provides a computer program including computer-readable code, wherein when the computer-readable code is run in a vehicle, a processor in the vehicle executes some or all of the steps in the above-described method.

[0145] This application provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program. When the computer program is read and executed by a computer, it implements some or all of the steps in the above-described method. This computer program product can be implemented specifically through hardware, software, or a combination thereof. In some embodiments, the computer program product is specifically embodied as a computer storage medium; in other embodiments, the computer program product is specifically embodied as a software product, such as a software development kit (SDK), etc.

[0146] It should be noted that the descriptions of the various embodiments above tend to emphasize the differences between them, while their similarities or commonalities can be referred to interchangeably. The descriptions of the above embodiments of the device, storage medium, computer program, and computer program product are similar to the descriptions of the above method embodiments and have similar beneficial effects. For technical details not disclosed in the embodiments of the device, storage medium, computer program, and computer program product of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0147] The aforementioned processor can be at least one of the following: Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), Controller, Microcontroller, and Microprocessor. It is understood that other electronic devices can also implement the functions of the aforementioned processor, and this application does not specifically limit the specific implementation.

[0148] The aforementioned computer storage media / memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), magnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM), etc.; or it can be various terminals that include one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc.

[0149] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above steps / processes do not imply a sequential order of execution; the execution order of each step / process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above embodiments of this application are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0150] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0151] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0152] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0153] In addition, each functional unit in the various embodiments of this application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0154] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.

[0155] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence or the part that contributes to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an in-vehicle terminal (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0156] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of this application are included within the scope of protection of this application.

Claims

1. A method for determining a drivable area, characterized in that, The method includes: The system acquires multi-source sensor data and the vehicle's driving environment around the vehicle; wherein the multi-source sensor data includes: visual spatial point cloud and / or target attribute information output by a visual sensor, and radar spatial point cloud output by a radar sensor, where the spatial points are the boundary points of the drivable area identified by the sensors. The multi-source sensor data is subjected to non-uniform polar coordinate sector grid projection and data structuring processing to obtain the first grid information corresponding to the same grid to which various spatial points belong; wherein, the first grid information includes at least one of the following: the polar radius of various spatial points under the non-uniform polar coordinate grid, sector identifier, distance band identifier under the sector to which the sector identifier belongs, sensor source identifier, first acquisition time, target attribute information, sensor basic reliability and historical stability count; Based on the vehicle driving environment and the first grid information corresponding to each grid, determine the conflict coding features corresponding to each type of spatial point within the grid; Obtain the baseline log confidence score for each grid cell. Based on the conflict coding features of various spatial points within each grid cell, the first grid cell information, and the baseline log confidence score, traverse all types of spatial points within the grid cell to perform multi-source data fusion and obtain the target log confidence score for the grid cell. For the same sector, the target log confidence of each grid is traversed radially from near to far. The polar radius of the first grid that meets the target log confidence condition is determined as the target polar radius of the sector. Based on the target polar radii of all sectors, the current drivable area of ​​the vehicle is determined.

2. The method according to claim 1, characterized in that, The non-uniform polar coordinate grid projection and data structuring processing of the multi-source sensor data includes: Based on the angular resolution of various radar sensors, and according to the preset projection rules, the visual space point cloud is subjected to non-uniform polar coordinate grid projection to determine the sector identifier of the grid to which each visual space point belongs, as well as the distance band identifier under the sector corresponding to the sector identifier. Based on the sector identifier and distance band identifier of the grid to which the visual spatial point belongs, the visual spatial point is subjected to unified structural processing to obtain the first grid information of the grid to which the visual spatial point belongs.

3. The method according to claim 1, characterized in that, The step of determining the conflict coding features corresponding to various spatial points within a grid based on the vehicle driving environment and the first grid information corresponding to each grid includes: For the radar spatial point cloud, the sector that meets the active conditions in the non-uniform coordinate sector is determined as the target sector; wherein, the active conditions include: the sector that has a valid observation of a visual point, a front radar blocking point, or a corner radar lateral point in the previous frame, and / or, the sector where the boundary spatial point of the previous frame fused output is located, and its left and right adjacent angular sectors. Based on the vehicle driving environment and the first grid information corresponding to the radar spatial points in each grid within the target sector, the collision coding features corresponding to the radar spatial points in the grid are determined.

4. The method according to claim 1, characterized in that, The step of determining the conflict coding features corresponding to various spatial points within the grid based on the vehicle driving environment and the first grid information corresponding to each grid includes at least one of the following: Based on the first grid information corresponding to the grid to which each visual spatial point belongs in the visual spatial point cloud and the historical frame polar radius corresponding to all sectors, the boundary gradient of adjacent range bands in the same sector and the polar radius jump variable of each grid visual spatial point in adjacent frames are determined. If the boundary gradient is greater than the gradient threshold, the polar radius jump variable is greater than the jump variable threshold, and there is no radar spatial point corresponding to a low target in the grid, it is determined that the visual spatial point in the grid has the visual near-range closed-radar long-range passable feature. Historical frame radar spatial point clouds are obtained. Based on the historical frame radar spatial point clouds and the radar spatial point clouds, radar grid information of each grid is statistically analyzed. For each grid, if the radar grid information satisfies both a first condition and a second condition, it is determined that the radar spatial points within the grid have consistency characteristics. If the radar grid information satisfies the first condition but not the second condition, it is determined that the radar spatial points within the grid have historical jump characteristics. The radar grid information includes the number of times radar boundary points appear when determining the drivable area corresponding to the historical frame, the polar radius variance, and the directional stability. The first condition includes: the number of occurrences is greater than a threshold value. The second condition includes: the polar radius variance is greater than a variance threshold value, and the directional stability is greater than a stability threshold value. Obtain the polar radius difference between visual spatial points and radar spatial points located in the same grid. If the polar radius difference is greater than the polar radius difference threshold, it is determined that the radar spatial points in the grid have radar near-range blocking and visual long-range passability characteristics. If the polar radius difference is less than the polar radius difference threshold, it is determined that the visual spatial points in the sector have visual near-range closure and radar long-range passability characteristics. When the vehicle is traveling in a tunnel environment, the tunnel area is determined based on high-precision map information and tunnel marker information output by visual sensors, and radar spatial points within the tunnel area are determined to have suspected tunnel wall features. Based on the forward-looking high-altitude marker information output by the visual sensor and the radar high-altitude marker position output by the radar sensor, the high-altitude region where the high-altitude target is located is determined, and the radar spatial points located within the high-altitude region are determined to have suspected high-altitude target features. Obtain the current frame system time. If the time difference between the current frame system time and the first acquisition time is greater than the time difference threshold, determine that various spatial points within the grid have time expiration characteristics.

5. The method according to any one of claims 1 to 4, characterized in that, The step of traversing all types of spatial points within the grid to perform multi-source data fusion, based on the conflict coding features corresponding to various types of spatial points in each grid, the first grid information, and the baseline log-sense confidence, to obtain the target log-sense confidence of the grid includes: Based on the current system time and the first acquisition time, determine the reliability of the timestamps corresponding to various spatial points; Based on the timestamp reliability, the conflict coding features, and the first grid information, the gating coefficients corresponding to each spatial point within the grid are determined. Obtain the logarithmic confidence increments corresponding to various spatial points within the grid; Based on the logarithmic confidence increment and gating coefficient corresponding to various spatial points within the grid, the baseline logarithmic confidence is updated in a fixed-point manner until all types of spatial points within the grid are traversed to achieve multi-source data fusion, thereby obtaining the target logarithmic confidence of the grid.

6. The method according to claim 5, characterized in that, The step of determining the gating coefficients corresponding to each spatial point within the grid based on the timestamp reliability, the conflict coding features, and the first grid information includes: Based on the vehicle driving environment, determine the basic reliability factor of various sensors at various spatial points within the grid; Based on the sector identifier and distance band identifier, determine the spatial distance factor of various spatial points within the grid; Obtain the conflict coding feature factor corresponding to the conflict coding feature, and the historical stability factor corresponding to the historical stability. Based on the timestamp reliability, sensor basic reliability factor, spatial distance factor, conflict coding feature factor, and historical stability factor, the gating coefficient corresponding to each spatial point within the grid is determined.

7. The method according to claim 5, characterized in that, The baseline logarithmic confidence level is updated using fixed-point updates based on the logarithmic confidence increments and gating coefficients corresponding to various spatial points within the raster, until all types of spatial points within the raster have been traversed to achieve multi-source data fusion, thereby obtaining the target logarithmic confidence level of the raster, including: If the grid includes visual spatial points and radar spatial points, the baseline logarithmic confidence of the grid is updated based on the logarithmic confidence increment and gating coefficient corresponding to the visual spatial points in the grid, to obtain the first intermediate logarithmic confidence of the grid. Based on the logarithmic confidence increment and gating coefficient corresponding to the radar spatial point within the grid, the first intermediate logarithmic confidence of the grid is updated to obtain the target logarithmic confidence.

8. The method according to any one of claims 1 to 4, characterized in that, Before performing non-uniform polar coordinate grid projection and data structuring processing on the multi-source sensor data to obtain the first grid information corresponding to the grid to which each spatial point belongs, the method includes: Obtain time-series vehicle status information; wherein, the time-series vehicle status information includes the second acquisition time, vehicle speed, and yaw rate under different frames; Starting from the second acquisition time of the spatial point cloud output by various sensors and ending with the current system time, a target time interval is determined, and the reference vehicle status information falling into the target time interval is traversed in the time-series vehicle status information. Based on the reference vehicle state information, the inter-frame difference of yaw rate between adjacent frames is determined, and based on the inter-frame difference of yaw rate, the target yaw rate increment is determined. Based on the target yaw rate increment and vehicle speed, motion compensation is performed on the corresponding sensor data to obtain compensated multi-source sensor data.

9. The method according to claim 8, characterized in that, Determining the target yaw rate increment based on the inter-frame difference in yaw rate includes: If the inter-frame difference in yaw rate is greater than the difference threshold, the yaw rate of the current frame is limited, and the limited yaw rate of the current frame is integrated to obtain the target yaw rate increment. If the inter-frame difference in yaw rate is less than or equal to the difference threshold, the yaw rate of the current frame is integrated to obtain the target yaw rate increment.

10. A device for determining a drivable area, characterized in that, include: Memory is used to store executable instructions or computer programs. A processor, when executing computer-executable instructions or computer programs stored in the memory, implements the drivable area determination method according to any one of claims 1 to 9.

Images