A camera-assisted laser dynamic point cloud culling method, device and storage medium

Abstract

The application discloses a camera-assisted laser dynamic point cloud elimination method and device and a storage medium. Firstly, a laser radar and a depth camera are jointly calibrated in space, and point clouds and visual features are obtained. Then, an improved random sample consensus (RANSAC) algorithm is used to eliminate outliers, and a locally iterative least squares (LILS) cycle is added to better improve the quality of the inlier set, so that a dynamic visual area is obtained. Through the correlation between the visual image and the point cloud, a dynamic point cloud area, hereinafter referred to as a dynamic background area, is divided. Further, the point cloud features with large residuals in the dynamic background area are regarded as dynamic point clouds to improve the precision of eliminating dynamic features. Then, based on region growing, dynamic targets are clustered, and finally, the dynamic targets are eliminated, thereby improving the positioning precision and mapping effect of SLAM. The application provides support for subsequent navigation and path planning of an unmanned vehicle.

Description

Technical Field

[0001] This invention relates to a camera-assisted laser dynamic point cloud culling method, device, and storage medium, belonging to the field of lidar data processing technology. Background Technology

[0002] Simultaneous Localization and Mapping (SLAM) is a good solution for autonomous vehicle navigation, providing map information and the vehicle's own position in unknown environments. However, many dynamic factors in the environment can violate the static assumptions of the SLAM algorithm, resulting in poor system robustness and low positioning accuracy. In real-world environments, moving pedestrians, vehicles, and other objects are unavoidable. Once these dynamic objects disappear or change position, they can significantly interfere with subsequent localization and attitude estimation, even causing SLAM mapping to fail. Therefore, eliminating dynamic factors from the environment is particularly important.

[0003] Removing dynamic objects is a crucial problem in the field of laser SLAM, and many scholars have conducted in-depth research on it. Wei et al. extracted vehicles from point clouds using adaptive 3D segmentation technology and estimated the vehicle's motion state and speed based on motion artifacts, achieving dynamic vehicle detection. However, the model built by this method can only identify dynamic vehicles and is mainly suitable for sparse point cloud datasets, performing poorly on dense point cloud datasets. Chen et al. reliably removed dynamic objects by constructing semantic maps and added semantic constraints to optimize scan matching, significantly improving the accuracy of laser odometry. However, this method requires large data storage space and prior map information, limiting its application. Zhang et al. proposed a pose estimation method based on dynamic feature removal and point cloud fusion with image fusion, effectively fusing pose results through an adaptive weighting strategy. Li et al. proposed a fully convolutional neural network that uses laser point clouds as input samples and combines data augmentation training to achieve dynamic target detection. The above method utilizes deep learning technology to effectively identify specific objects in images. By combining prior knowledge of moving objects (cars, pedestrians, animals), it can effectively remove potential dynamic regions. Experimental results demonstrate that this method is the simplest and most efficient solution, and it is also the current mainstream technical solution.

[0004] However, it should be noted that existing methods often make incorrect identifications for some non-mainstream moving objects, such as mobile stalls parked on the roadside, cars, and shared bicycles. Furthermore, the dynamic point cloud removal methods described above do not consider the mismatch problem caused by the presence of outliers.

[0005] Therefore, in order to ensure the accuracy of dynamic point cloud removal, those skilled in the art urgently need to improve the existing laser dynamic point cloud removal methods. Summary of the Invention

[0006] Objective: To overcome the problem that many dynamic factors in the environment violate the static assumptions of the SLAM algorithm, resulting in poor system robustness and low positioning accuracy, the present invention provides a camera-assisted laser dynamic point cloud culling method, device and storage medium. It not only eliminates the interference of outliers on dynamic point cloud culling, but also uses a clustering method of dynamic targets based on residuals and region growing to culle dynamic point cloud features, avoiding the interference of "ghosting" on SLAM localization and mapping.

[0007] Technical solution: To solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0008] Firstly, a camera-assisted laser dynamic point cloud culling method includes the following steps:

[0009] Step 1: Use a LiDAR scanner to scan non-standard moving objects and acquire LiDAR 3D point cloud data. Then, use a camera to photograph these non-standard moving objects and acquire camera 3D point cloud data.

[0010] Step 2: Convert the LiDAR 3D point cloud data to the camera coordinate system to obtain the converted LiDAR 3D point cloud data. Then, fuse the converted LiDAR 3D point cloud data with the camera 3D point cloud data to obtain the fused 3D point cloud data.

[0011] Step 3: Project the fused 3D point cloud data onto the image plane according to the camera intrinsic parameters to obtain the fused 2D point cloud data.

[0012] Step 4: Randomly select n data points (n≥3) from the fused two-dimensional point cloud data, and fit the curve y(q) using the least squares method based on the n data points.

[0013] Step 5: Calculate any point q in the fused 2D point cloud data j The distance to curve f(q), when the distance is less than a threshold, for any point q j For any point q, it is an interior point, and when the distance is greater than or equal to a threshold. j It belongs to the outer point, and records the number of all inner points.

[0014] Step 6: Repeat steps 4 to 5 until the number of iterations is completed, and output the outer point corresponding to the maximum number of inner points.

[0015] Step 7: Remove the outliers from the fused 2D point cloud data to obtain the dynamic point cloud region.

[0016] Step 8: Obtain the dynamic point cloud region of the current frame and the dynamic point cloud region of the previous frame. Calculate the moving distance of each point in the dynamic point cloud region between the current frame and the previous frame. Compare the moving distance with the moving threshold. Remove the points in the dynamic point cloud region of the current frame that are less than the moving threshold to obtain the optimized dynamic point cloud region of the current frame.

[0017] Step 9: Use the region growing method to cluster the optimized current frame dynamic point cloud region, and use the resulting clustered region as the final dynamic point cloud region.

[0018] Step 10: Perform an inverse coordinate transformation on the final dynamic point cloud region to obtain the final laser dynamic point cloud data.

[0019] As a preferred solution, the 3D point cloud in the LiDAR coordinate system is represented as follows:

[0020]

[0021] in, This represents the i-th point cloud in the LiDAR coordinate system. Y i L , Let X and Y represent the coordinates of the i-th point cloud in the LiDAR coordinate system along the X, Y, and Z axes, respectively, and T denotes transpose.

[0022] The point cloud representation in the Camera coordinate system is as follows:

[0023]

[0024] in, This represents the i-th point cloud in the Camera coordinate system. Y i C , Let X and Y represent the coordinates of the i-th point cloud in the X, Y, and Z axes of the Camera coordinate system, respectively, and T denotes transpose.

[0025] As a preferred solution, the laser point cloud is transformed into the camera coordinate system using the following transformation:

[0026]

[0027] in, Let R represent the i-th point cloud of the transformed lidar, R represent the rotation matrix, and t represent the translation matrix.

[0028] As a preferred embodiment, the process of fusing the converted lidar 3D point cloud data with the camera 3D point cloud data to obtain fused 3D point cloud data specifically includes:

[0029] One of the point cloud data at the same location between the converted LiDAR 3D point cloud data and the camera 3D point cloud data is retained.

[0030] The point cloud data at different locations in the converted LiDAR 3D point cloud data and camera 3D point cloud data are all preserved.

[0031] The retained point cloud data is used as the fused 3D point cloud data.

[0032] As a preferred approach, the points in three-dimensional space are projected onto a two-dimensional image plane, and the calculation formula is as follows:

[0033]

[0034] in, k represents the fused 3D point cloud data. c q represents the camera intrinsic parameter function. i This represents the fused two-dimensional point cloud data.

[0035] As a preferred option, the formula for calculating curve y(q) is as follows:

[0036] y(q)=a0+a1x+a2x 2 +……+a k x k

[0037] Among them, a k Let x represent the k-th undetermined coefficient, where x is the value of the x-axis of the curve y(q), and k < n.

[0038] As a preferred embodiment, the threshold is a continuously adjusted threshold t', and the calculation formula for the threshold t' is as follows:

[0039] t′=t0+(t1-t0)*k t

[0040] Where t0 is the set initial threshold, t1 is the set target threshold, and k t This represents the proportion of the current threshold t during the adjustment process.

[0041] As a preferred embodiment, step 9 specifically includes:

[0042] Step 9.1: Take any pixel in the optimized current frame dynamic point cloud region as a seed point and set the growth criteria.

[0043] Step 9.2: Determine whether the neighboring points near the seed point meet the growth criteria. If they do, add the neighboring points to the clustering region; otherwise, mark the neighboring points as visited points.

[0044] Step 9.3: Repeat steps 9.1 to 9.2 until all neighboring points are added to the cluster region or marked as visited points, and output the cluster region as the final dynamic point cloud region.

[0045] In a second aspect, a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a camera-assisted laser dynamic point cloud culling method as described in any of the first aspects.

[0046] Thirdly, a computer device comprising:

[0047] Memory is used to store instructions.

[0048] A processor is configured to execute the instructions, causing the computer device to perform operations of a camera-assisted laser dynamic point cloud culling method as described in any of the first aspects.

[0049] Beneficial Effects: This invention provides a camera-assisted laser dynamic point cloud culling method, device, and storage medium. First, spatial joint calibration of the LiDAR and depth camera is performed, simultaneously obtaining point cloud and visual features. Then, an improved Random Sample Consensus (RANSAC) algorithm is used to culle outliers, incorporating a Locally Iterative Least Squares (LILS) loop to further improve the quality of the inlier set, thereby obtaining a dynamic visual region. By associating the visual image with the point cloud, a dynamic point cloud region, hereinafter referred to as the dynamic background region, is delineated. Furthermore, point cloud features with large residuals within the dynamic background region are considered as dynamic point clouds to improve the accuracy of dynamic feature culling. Then, dynamic target clustering is performed based on region growing, and finally, these clusters are culled, thereby improving the localization accuracy and mapping effect of SLAM, providing support for subsequent navigation and path planning operations of unmanned vehicles. Compared with existing technologies, its beneficial effects are as follows:

[0050] (1) The SensorCalibration Toolbox is used for the spatial joint calibration of the LiDAR and the camera. The calibration process is to adjust the internal and external parameters to align the LiDAR point cloud and the image in the road scene, so that the point cloud and visual features can be obtained at the same time.

[0051] (2) An improved RANSAC algorithm was used to remove outliers by incorporating a local iterative least squares loop, which improved the quality of the inlier set and obtained dynamic visual regions, thereby reducing mismatches in the feature matching process. Dynamic point cloud regions were delineated by associating visual images with point clouds. Furthermore, point cloud features with large residuals within these regions were considered dynamic point clouds, and then a region-growing-based method was used to cluster dynamic targets and remove them. This improved the accuracy of dynamic feature removal, thereby enhancing the localization accuracy and mapping performance of SLAM. Attached Figure Description

[0052] Figure 1 This is a flowchart of a camera-assisted laser dynamic point cloud culling method. Detailed Implementation

[0053] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0054] The present invention will be further described below with reference to specific embodiments.

[0055] Example 1:

[0056] This embodiment describes a camera-assisted laser dynamic point cloud culling method, including the following steps:

[0057] Step 1: Use a LiDAR scanner to scan non-standard moving objects and acquire LiDAR 3D point cloud data. Then, use a camera to photograph these non-standard moving objects and acquire camera 3D point cloud data.

[0058] Furthermore, the 3D point cloud representation in the LiDAR coordinate system is as follows:

[0059]

[0060] in, This represents the i-th point cloud in the LiDAR coordinate system. Y i L , Let X and Y represent the coordinates of the i-th point cloud in the LiDAR coordinate system along the X, Y, and Z axes, respectively, and T denotes transpose.

[0061] Furthermore, the point cloud representation in the Camera coordinate system is as follows:

[0062]

[0063] in, This represents the i-th point cloud in the Camera coordinate system. Y i C , Let X and Y represent the coordinates of the i-th point cloud in the X, Y, and Z axes of the Camera coordinate system, respectively, and T denotes transpose.

[0064] Step 2: Convert the LiDAR 3D point cloud data to the camera coordinate system to obtain the converted LiDAR 3D point cloud data. Then, fuse the converted LiDAR 3D point cloud data with the camera 3D point cloud data to obtain the fused 3D point cloud data.

[0065] Furthermore, the laser point cloud is transformed into the camera coordinate system using the following transformation:

[0066]

[0067] in, Let R represent the i-th point cloud of the transformed lidar, R represent the rotation matrix, and t represent the translation matrix, which together constitute the extrinsic parameters.

[0068] Furthermore, the process of fusing the converted lidar 3D point cloud data with the camera 3D point cloud data to obtain fused 3D point cloud data specifically includes:

[0069] One of the point cloud data at the same location between the converted LiDAR 3D point cloud data and the camera 3D point cloud data is retained.

[0070] The point cloud data at different locations in the converted LiDAR 3D point cloud data and camera 3D point cloud data are all preserved.

[0071] The retained point cloud data is used as the fused 3D point cloud data.

[0072] Step 3: Project the fused 3D point cloud data onto the image plane according to the camera intrinsic parameters to obtain the fused 2D point cloud data.

[0073] Furthermore, this process requires using the camera's intrinsic parameters to project points in three-dimensional space onto a two-dimensional image plane. The calculation formula is as follows:

[0074]

[0075] in, k represents the fused 3D point cloud data. c q represents the camera intrinsic parameter function. i This represents the fused two-dimensional point cloud data.

[0076] Because different sensors use different coordinate systems, the data collected by various sensors for the same object will differ. Therefore, joint calibration of LiDAR and camera is necessary. This paper uses the SensorCalibration Toolbox for spatial joint calibration of LiDAR and camera. The calibration process involves adjusting intrinsic and extrinsic parameters to align the LiDAR point cloud and image point cloud of non-mainstream moving objects in the road scene.

[0077] Step 4: Randomly select n data points (n≥3) from the fused two-dimensional point cloud data, and fit the curve y(q) using the least squares method based on the n data points.

[0078] Furthermore, the formula for calculating the curve y(q) is as follows:

[0079] y(q)=a0+a1x+a2x 2 +……+a k x k

[0080] Among them, a k Let x represent the k-th undetermined coefficient, where x is the value of the x-axis of the curve y(q), and k < n.

[0081] Step 5: Calculate any point q in the fused 2D point cloud data j The distance to curve f(q), when the distance is less than a threshold, for any point q j For any point q, it is an interior point, and when the distance is greater than or equal to a threshold. j It belongs to the outer point, and records the number of all inner points.

[0082] Furthermore, the threshold is a key parameter for determining the degree of matching between data and the model. If the threshold t is too large, some inliers that should be included in the model may be treated as outliers, thus affecting the accuracy of the model. Conversely, if the threshold t is too small, some noisy data may be mistakenly included in the model, which will also reduce the accuracy of the model. This invention is aimed at large-scale field scenarios containing a large number of outliers, which are prone to mismatches. Therefore, the initial value of the threshold t needs to be set relatively large to prevent outliers from being included in the model. The threshold t' is continuously adjusted using a linear interpolation method.

[0083] t′=t0+(t1-t0)*k t (6)

[0084] Where t0 is the set initial threshold, t1 is the set target threshold, and k t This represents the proportion of the current threshold t during the adjustment process.

[0085] Step 6: Repeat steps 4 to 5 until the number of iterations is completed, and output the outer point corresponding to the maximum number of inner points.

[0086] Step 7: Remove the outliers from the fused 2D point cloud data to obtain the dynamic point cloud region.

[0087] A local iterative least squares loop is introduced to improve the RANSAC algorithm for outlier removal, thus improving the quality of the inlier set and obtaining a dynamic visual region. Through iteration, RANSAC can perform local optimization in each iteration, improving the accuracy of model fitting and thereby removing outliers more accurately.

[0088] By jointly calibrating the spatial data using LiDAR and cameras, and associating visual and point cloud features, the dynamic point cloud region, i.e., the dynamic background region, is obtained after removing outliers. This process reduces mismatches by removing dynamic point clouds and prevents outliers from being incorrectly classified as dynamic points, thereby improving the localization and mapping accuracy of SLAM.

[0089] Step 8: Obtain the dynamic point cloud region of the current frame and the dynamic point cloud region of the previous frame. Calculate the moving distance of each point in the dynamic point cloud region between the current frame and the previous frame. Compare the moving distance with the moving threshold. Remove the points in the dynamic point cloud region of the current frame that are less than the moving threshold to obtain the optimized dynamic point cloud region of the current frame.

[0090] Step 9: Use the region growing method to cluster the optimized current frame dynamic point cloud region, and use the resulting clustered region as the final dynamic point cloud region.

[0091] Furthermore, step 9 specifically includes:

[0092] Step 9.1: Take any pixel in the optimized current frame dynamic point cloud region as a seed point and set the growth criteria.

[0093] Step 9.2: Determine whether the neighboring points near the seed point meet the growth criteria. If they do, add the neighboring points to the clustering region; otherwise, mark the neighboring points as visited points.

[0094] Step 9.3: Repeat steps 9.1 to 9.2 until all neighboring points are added to the cluster region or marked as visited points, and output the cluster region as the final dynamic point cloud region.

[0095] Step 10: Perform an inverse coordinate transformation on the final dynamic point cloud region to obtain the final laser dynamic point cloud data.

[0096] At this point, potential dynamic and static points are initially distinguished. Then, by combining region growing methods, dynamic targets are effectively clustered and eliminated to avoid "ghosting." This also ensures the positioning accuracy and mapping effect of the SLAM system.

[0097] Example 2:

[0098] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a camera-assisted laser dynamic point cloud culling method as described in any of Embodiment 1.

[0099] Example 3:

[0100] A computer device, comprising:

[0101] Memory is used to store instructions.

[0102] A processor is configured to execute the instructions, causing the computer device to perform operations of a camera-assisted laser dynamic point cloud culling method as described in any of Embodiment 1.

[0103] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0104] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0105] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0106] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0107] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A camera-assisted laser dynamic point cloud removal method, characterized in that: Includes the following steps: Step 1: Use a lidar to scan non-mainstream moving objects and obtain lidar 3D point cloud data; use a camera to photograph non-mainstream moving objects and obtain camera 3D point cloud data. Step 2: Convert the LiDAR 3D point cloud data to the camera coordinate system to obtain the converted LiDAR 3D point cloud data. Then, fuse the converted LiDAR 3D point cloud data with the camera 3D point cloud data to obtain the fused 3D point cloud data. Step 3: Project the fused 3D point cloud data onto the image plane according to the camera intrinsic parameters to obtain the fused 2D point cloud data; Step 4: Randomly select from the fused 2D point cloud data Data points, ,according to Using a set of data points, the curve is fitted using the least squares method. ; Step 5: Calculate any point in the fused 2D point cloud data to curve The distance, when the distance is less than the threshold, for any point If a point is an interior point, and the distance is greater than or equal to a threshold, then any point... It belongs to the outer point, and records the number of all inner points; Step 6: Repeat steps 4 to 5 until the number of iterations is completed, and output the outer point corresponding to the maximum number of inner points; Step 7: Remove the outliers from the fused 2D point cloud data to obtain the dynamic point cloud region; Step 8: Obtain the dynamic point cloud region of the current frame and the dynamic point cloud region of the previous frame. Calculate the moving distance of each point in the dynamic point cloud region between the current frame and the previous frame. Compare the moving distance with the moving threshold. Remove the points in the dynamic point cloud region of the current frame that are less than the moving threshold to obtain the optimized dynamic point cloud region of the current frame. Step 9: Cluster the optimized current frame dynamic point cloud region using the region growing method, and use the resulting clustered region as the final dynamic point cloud region. Step 10: Perform an inverse coordinate transformation on the final dynamic point cloud region to obtain the final laser dynamic point cloud data; The threshold is a continuously adjusted threshold t', and the formula for calculating the threshold t' is as follows: ； in, The initial threshold is set. For the set target threshold, This represents the proportion of the current threshold t during the adjustment process.

2. The camera-assisted laser dynamic point cloud removal method according to claim 1, characterized in that: The 3D point cloud representation in LiDAR coordinates is as follows: ； in, Represents the first in Lidar coordinate system A point cloud, They represent the first The coordinate values ​​of a point cloud in the X, Y, and Z axes of the Lidar coordinate system, where T represents transpose; The point cloud representation in the camera coordinate system is as follows: ； in, Represents the first in the camera coordinate system A point cloud, They represent the first The coordinates of a point cloud along the X, Y, and Z axes in the camera coordinate system, where T represents transpose.

3. The camera-assisted laser dynamic point cloud removal method according to claim 1, characterized in that: The laser point cloud is transformed into the camera coordinate system using the following transformation: ； in, The first laser radar after conversion A point cloud, where R represents the rotation matrix and t represents the translation matrix. Represents the first in Lidar coordinate system Point cloud.

4. The camera-assisted laser dynamic point cloud removal method according to claim 1, characterized in that: The process of fusing the converted lidar 3D point cloud data with the camera 3D point cloud data to obtain fused 3D point cloud data specifically includes: One of the point cloud data at the same location between the converted LiDAR 3D point cloud data and the camera 3D point cloud data is retained; The point cloud data at different locations in the converted LiDAR 3D point cloud data and camera 3D point cloud data are all preserved; The retained point cloud data is used as the fused 3D point cloud data.

5. The camera-assisted laser dynamic point cloud removal method according to claim 1, characterized in that: The formula for projecting a point in three-dimensional space onto a two-dimensional image plane is as follows: ； in, This represents the fused 3D point cloud data. This represents the camera intrinsic parameter function. This represents the fused two-dimensional point cloud data.

6. The camera-assisted laser dynamic point cloud removal method according to claim 1, characterized in that: Step 9 specifically includes: Step 9.1: Take any pixel in the optimized current frame dynamic point cloud region as a seed point and set the growth criteria; Step 9.2: Determine whether the neighboring points near the seed point meet the growth criteria. If they do, add the neighboring points to the clustering region; otherwise, mark the neighboring points as visited points. Step 9.3: Repeat steps 9.1 to 9.2 until all neighboring points are added to the cluster region or marked as visited points, and output the cluster region as the final dynamic point cloud region.

7. The camera-assisted laser dynamic point cloud removal method according to claim 1, characterized in that: curve The calculation formula is as follows: ； in, Indicates the first One undetermined coefficient For curves of The value of the axis. , This represents the number of data points.

8. A computer-readable storage medium, characterized in that: It stores a computer program that, when executed by a processor, implements a camera-assisted laser dynamic point cloud removal method as described in any one of claims 1-7.

9. A computer device, characterized in that: include: Memory, used to store instructions; A processor is configured to execute the instructions, causing the computer device to perform the operation of a camera-assisted laser dynamic point cloud culling method as described in any one of claims 1-7.

Images