BIM-based modular hoisting path planning and collision detection system for a mixing plant

Abstract

The application discloses a BIM-based mixing station modular hoisting path planning and collision detection system and belongs to the technical field of building construction informatization. In view of the problems of the conflict between detection accuracy and efficiency caused by the great difference in geometric shapes of modules and the high missing detection rate of fixed steps caused by uneven distribution of obstacles, the application determines the heterogeneous or regular modules by the size ratio of the outer-approximate orthogonal cuboid of the modules; adopts an optimized octree algorithm to divide the hoisting operation space and record the node obstacle density; adopts a mixed bounding box combined with orientation and axis alignment for the heterogeneous modules and adopts an axis-aligned bounding box for the regular modules; and when performing collision detection on the initial path, according to the node obstacle density and the minimum distance between the module and the nearest obstacle, the first or second detection step is dynamically adopted, intersection test is performed and the path is optimized. The system is used for path planning and collision detection of the mixing station modular hoisting, can reduce the calculation amount under the guarantee of accuracy and realizes fast and accurate planning.

Description

Technical Field

[0001] This invention belongs to the field of information technology in building construction, specifically relating to a BIM-based modular hoisting path planning and collision detection system for mixing plants. Background Technology

[0002] The modular construction method of the mixing plant involves disassembling the mixing host, batching machine, powder tank, control room, etc., into independent modules, which are prefabricated in the factory and then transported to the site for hoisting and assembly. The geometric shapes of each module vary significantly; some are slender rods, some are flat boxes, and some are regular structures that are approximately cubic. Before the hoisting operation, it is necessary to plan the spatial movement path of the lifting equipment to move the module from the lifting point to the target installation location. During the planning process, potential collisions between the module and surrounding installed structures, temporary stacks, and the lifting equipment itself must be pre-detected and avoided.

[0003] Existing collision detection methods typically employ a single bounding box approach to geometrically approximate the lifting modules, such as uniformly using axis-aligned bounding boxes or uniformly using oriented bounding boxes. Axial-aligned bounding boxes offer less tight coverage for slender or irregularly shaped modules, containing significant amounts of blank space outside the module itself. This leads to false collision alarms in narrow passageways, forcing path planning algorithms to detour or repeatedly adjust, increasing the computational cost of path search. While oriented bounding boxes provide a closer fit to the module's geometry, their intersection testing computation is significantly higher than that of axis-aligned bounding boxes. Applying oriented bounding boxes indiscriminately to all modules accumulates a considerable computational burden throughout the path detection process, impacting overall planning efficiency. The diverse geometric dimensions of mixing plant modules make it difficult to strike a balance between detection accuracy and computational efficiency using a single bounding box strategy.

[0004] Furthermore, the construction site of the mixing plant has a compact spatial layout. The hoisting operation space contains both installed structural components and temporarily stored building materials and equipment, resulting in an uneven distribution of obstacles. Existing path discretization collision detection methods often use a fixed step size for sampling the path. If the step size is set too large, in areas with dense obstacles or path segments where modules are close to obstacles, critical collision points may fall between two sampling points and be missed, leading to a risk of missed detections. If the step size is set too small, a large number of redundant detection points will be generated in wide areas with sparse obstacles, resulting in unnecessary intersection test calculations and prolonging path planning time. How to adaptively adjust the detection density to address the uneven distribution of obstacles within the hoisting space, and control the overall number of detections while avoiding missed detections, is a key challenge in improving the reliability and efficiency of collision detection. Summary of the Invention

[0005] One object of the present invention is to solve at least the above-mentioned problems and to provide at least the advantages that will be described later.

[0006] Another objective of this invention is to provide a BIM-based modular hoisting path planning and collision detection system for mixing plants. This system can adaptively select the bounding box accuracy based on the geometric dimensions of the mixing plant modules, and dynamically adjust the detection step size according to the density and spacing of obstacles during the collision detection process of the hoisting path. This effectively controls the overall computational overhead while ensuring the reliability of the detection, and achieves hoisting path planning that balances accuracy and efficiency.

[0007] To achieve these objectives and other advantages of the present invention, a BIM-based modular hoisting path planning and collision detection system for mixing plants is provided. The construction of this system includes the following steps: BIM software was used to construct a three-dimensional construction scene model that includes the mixing plant module to be hoisted, hoisting equipment, installed structures, and obstacles on the construction site. The geometric dimensions of the circumscribed orthogonal cuboid of the batching plant module to be hoisted are obtained from the 3D construction scene model. When the ratio of the maximum dimension to the minimum dimension in the geometric dimension parameters exceeds the preset ratio threshold, the batching plant module to be hoisted is determined to be a heterogeneous module. When the ratio of the maximum dimension to the minimum dimension in the geometric dimension parameters does not exceed the preset ratio threshold, the batching plant module to be hoisted is determined to be a regular module. Based on a 3D construction scene model, the reachable space from the lifting point to the target installation position of the mixing plant module to be hoisted is defined as the hoisting operation space. An optimized octree algorithm is used to spatially divide and encode the hoisting operation space, generating an octree spatial structure. The optimized octree algorithm performs deep subdivision on areas where the obstacle distribution density is higher than a preset depth subdivision threshold, and shallow subdivision or no subdivision is used on areas where the obstacle distribution density does not exceed the preset depth subdivision threshold. During the subdivision process, the obstacle distribution density value within the space of each octree node is recorded. Obstacles include installed structures, construction site obstacles, and fixed structures of the hoisting equipment. Heterogeneous modules are geometrically enclosed using a hybrid bounding box combining directional bounding boxes and axis-aligned bounding boxes, while regular modules are geometrically enclosed using axis-aligned bounding boxes. The hybrid bounding box parameters of the heterogeneous modules and the axis-aligned bounding box parameters of the regular modules can be synchronously updated in real time according to the real-time lifting posture of the modules corresponding to each detection position on the lifting path. At the same time, all obstacles and moving parts of the lifting equipment within the octree spatial structure are geometrically enclosed using axis-aligned bounding boxes, wherein the bounding box parameters of the moving parts of the lifting equipment can be synchronously updated in real time according to the real-time posture of the equipment corresponding to each detection position on the lifting path. Based on a 3D construction scene model and an octree spatial structure, an initial hoisting path is generated for the mixing plant module to be hoisted from the lifting point to the target installation position. The initial hoisting path is discretized to obtain several discrete path points. The preset discretization step size used in the discretization process is smaller than the second detection step size and not greater than the side length of the smallest leaf node in the octree spatial structure, ensuring that the spatial distance between any two adjacent discrete path points does not exceed the side length of the smallest leaf node. During the collision detection and optimization of the initial hoisting path, for the path segment between two adjacent discrete path points, the minimum spatial distance between the bounding box of the mixing plant module to be hoisted and the bounding box of the nearest obstacle in the path segment is calculated based on the octree spatial structure. When the obstacle density of any octree node traversed by the path segment is higher than a preset depth subdivision threshold, or the minimum spatial distance is less than a preset distance threshold, the first detection step size is used. When the obstacle density of all octree nodes traversed by the path segment does not exceed the preset depth subdivision threshold, and the minimum spatial distance is not less than the preset distance threshold, the second detection step size is used. The first detection step size is smaller than the second detection step size. According to the adjusted collision detection step size, at the corresponding detection position within each path segment, the bounding box parameters of the moving parts of the hoisting equipment are first updated based on the real-time attitude of the hoisting equipment at that detection position. Then, the bounding box of the hoisting mixing station module and the bounding box of the hoisting equipment's moving parts are simultaneously subjected to an intersection test with the bounding boxes of obstacles in the corresponding area of ​​the octree spatial structure to determine whether a collision exists. If a collision is detected, the discrete path points at the corresponding positions are adjusted, and the collision detection step size adjustment and intersection test are re-executed on the adjusted path segment until a collision-free hoisting path is obtained.

[0008] Preferably, a 3D construction scene model is constructed using BIM software, including the batching plant module to be hoisted, hoisting equipment, installed structures, and obstacles on the construction site. This model further includes the following dynamic update steps, which are performed before the hoisting operation of each batching plant module: At least one 3D laser scanner is deployed at the construction site of the mixing plant. Each 3D laser scanner collects point cloud data of the surrounding environment at different locations on the construction site. The point cloud data collected by each 3D laser scanner is registered and fused to generate a 3D point cloud model of the construction site covering the hoisting operation space; Extract the actual position coordinates and actual geometric dimensions of the installed structures and the actual position coordinates and actual geometric dimensions of obstacles on the construction site from the 3D point cloud model of the construction site. The actual position coordinates and actual geometric dimensions of the installed structure are extracted and compared with the design position coordinates and design geometric dimensions of the corresponding components in the BIM design model of the installed structure pre-built based on the design drawings. When the Euclidean distance between the actual position coordinates and the design position coordinates of a component exceeds the preset position tolerance threshold, the position coordinates of the component in the 3D construction scene model are updated to the actual position coordinates. When the difference between the actual geometric dimensions and the design geometric dimensions of a component exceeds the preset dimension tolerance threshold, the geometric dimensions of the component in the 3D construction scene model are updated to the actual geometric dimensions. The actual position coordinates and actual geometric dimensions of the extracted construction site obstacles are compared with the position coordinates and geometric dimensions of the construction site obstacles already recorded in the 3D construction scene model. When there is no obstacle record at the corresponding position of the actual position coordinates of the construction site obstacle in the 3D construction scene model, the construction site obstacle is added to the 3D construction scene model. When the position coordinates of a construction site obstacle recorded in the 3D construction scene model are not detected at the corresponding position in the 3D point cloud model of the construction site, the construction site obstacle is removed from the 3D construction scene model. After updating the position coordinates and geometric dimensions of the installed structure, adding obstacles to the construction site, and removing obstacles from the construction site, the steps of delineating the hoisting operation space based on the 3D construction scene model and using an optimized octree algorithm to divide and encode the hoisting operation space are repeated to generate an updated octree spatial structure. Simultaneously, based on the updated 3D construction scene model, the steps of determining the type of the mixing plant module to be hoisted, generating the bounding box of the mixing plant module to be hoisted, and generating bounding boxes of all obstacles, fixed structures and moving parts of the hoisting equipment within the octree spatial structure are repeated.

[0009] Preferably, the point cloud data collected by each 3D laser scanner is registered and fused to generate a 3D point cloud model of the construction site covering the hoisting operation space, further including: In the point cloud data collected by each 3D laser scanner, the spherical point cloud region generated by multiple target spheres deployed at the construction site is identified and extracted. Multiple target spheres are deployed at the boundary of the hoisting operation space and in the overlapping area of ​​the scanning range of each 3D laser scanner. The coordinates of the center of each target sphere in the coordinate system of the construction site are determined in advance by a total station. For each pair of 3D laser scanners with overlapping scanning areas, based on the spherical point cloud regions corresponding to at least three non-collinear target spheres in the point cloud data of each pair of 3D laser scanners, the least squares spherical fitting algorithm is used to calculate the first center coordinates and the second center coordinates of the target sphere in the corresponding scanner coordinate system. Based on the first center coordinates, the second center coordinates, and the pre-determined coordinates of the target sphere's center in the construction site coordinate system, a spatial transformation matrix between the pair of 3D laser scanners is constructed. The point cloud data collected by each 3D laser scanner is transformed point by point using a spatial transformation matrix, and the point cloud data of each 3D laser scanner is uniformly transformed to the construction site coordinate system to obtain the initial registration point cloud. In the initial registration point cloud, a key fusion area is defined for the spatial range corresponding to the vertical lifting channel and the horizontal turning channel that the mixing plant module to be lifted must pass through from the lifting point to the target installation position within the lifting operation space. The height range of the key fusion area in the vertical direction is from the first preset height above the plane where the lifting point is located to the second preset height above the plane where the target installation position is located. The range in the horizontal direction covers the shortest straight line connecting the lifting point and the target installation position and extends to both sides by a preset width range. Within the key fusion area, an adaptive weighted fusion algorithm based on the angle between normal vectors and curvature changes is used to fuse the overlapping point clouds from different 3D laser scanners. In the non-key fusion area outside the key fusion area, a uniform weighted fusion algorithm based on the average point spacing is used to fuse the overlapping point clouds from different 3D laser scanners. After the fusion process is completed, the point cloud voids located in the key fusion area in the 3D point cloud model of the construction site are filled using a point cloud void repair algorithm based on moving least squares surface.

[0010] Preferably, within the key fusion region, an adaptive weighted fusion algorithm based on the angle between normal vectors and curvature variations is used to fuse overlapping point clouds from different 3D laser scanners, further including: Within the key fusion area, if there is an overlap between a first point cloud from the first 3D laser scanner and a second point cloud from the second 3D laser scanner at any spatial location, a first neighborhood sphere is constructed with the first point cloud as the center and a preset neighborhood radius as the radius, and a second neighborhood sphere is constructed with the second point cloud as the center and a preset neighborhood radius as the radius. Calculate the first average normal vector and the first average curvature value of the local point cloud surface formed by all point cloud points in the first neighborhood sphere, and the second average normal vector and the second average curvature value of the local point cloud surface formed by all point cloud points in the second neighborhood sphere. Calculate the angle between the first average normal vector and the second average normal vector, and the absolute value of the difference between the first average curvature value and the second average curvature value; Based on the absolute values ​​of the included angle and the difference, the fusion weight coefficients of the first point cloud and the second point cloud are determined respectively. When the included angle exceeds a preset included angle threshold, the fusion weight coefficients of the first point cloud and the second point cloud are set as the first set of weight values. When the included angle does not exceed the preset included angle threshold and the absolute value of the difference exceeds a preset curvature difference threshold, the fusion weight coefficients of the first point cloud and the second point cloud are set as the second set of weight values. When the included angle does not exceed the preset included angle threshold and the absolute value of the difference does not exceed the preset curvature difference threshold, the fusion weight coefficients of the first point cloud and the second point cloud are set as the third set of weight values. The smaller weight value in the first set of weight values ​​is smaller than the smaller weight value in the second set of weight values, and the smaller weight value in the second set of weight values ​​is smaller than the smaller weight value in the third set of weight values. The sum of the fusion weight coefficients of the first point cloud and the second point cloud is one. By using the fusion weighting coefficients of the first and second point cloud points, a weighted average is performed on the spatial coordinates of the first and second point cloud points to obtain the spatial coordinates of the fused point cloud points.

[0011] Preferably, in non-key fusion areas outside the key fusion area, a uniform weighted fusion algorithm based on the mean point spacing is used to fuse overlapping point clouds from different 3D laser scanners, further including: In non-key fusion areas, if there is an overlap between a first point cloud from the first 3D laser scanner and a second point cloud from the second 3D laser scanner at any spatial location, a first search sphere is constructed with the first point cloud as the center and a preset search radius as the radius, and a second search sphere is constructed with the second point cloud as the center and a preset search radius as the radius. The average distance between all point cloud points within the first search sphere is calculated as the first local point cloud density index, and the average distance between all point cloud points within the second search sphere is calculated as the second local point cloud density index. The ratio of the first local point cloud density index to the second local point cloud density index is used as the point cloud density ratio, and the reciprocal of the ratio of the nominal ranging accuracy of the three-dimensional laser scanner corresponding to the first point cloud point and the second point cloud point is used as the scanner accuracy weighting factor. Based on the point cloud density ratio and the scanner accuracy weight factor, the first fusion weight coefficient of the first point cloud point and the second fusion weight coefficient of the second point cloud point are determined. The first fusion weight coefficient is negatively correlated with the first local point cloud density index and positively correlated with the nominal ranging accuracy of the scanner corresponding to the first point cloud point. The second fusion weight coefficient is negatively correlated with the second local point cloud density index and positively correlated with the nominal ranging accuracy of the scanner corresponding to the second point cloud point. The sum of the first fusion weight coefficient and the second fusion weight coefficient is one. The spatial coordinates of the first point cloud points and the second point cloud points are weighted and averaged using the first fusion weight coefficient and the second fusion weight coefficient to obtain the spatial coordinates of the fused point cloud points.

[0012] Preferably, based on a 3D construction scene model and an octree spatial structure, an initial hoisting path is generated for the batching plant module to be hoisted from the lifting point to the target installation position. The initial hoisting path is then discretized to obtain several discrete path points, further including: In the octree space structure, identify and mark all obstacle nodes that intersect with the bounding box of obstacles, and treat octree nodes that are not marked as obstacle nodes as free space nodes; Starting with the free space node where the lifting point is located as the starting node and the free space node where the target installation location is located as the target node, an improved fast expanding random tree algorithm is used to search for an initial path in the space formed by the free space nodes. When generating a new node through random sampling, the improved fast expanding random tree algorithm calculates the change in luffing angle and slewing angle of the lifting equipment boom corresponding to the expansion from the current node to the new node. When the change in luffing angle exceeds the preset single-step luffing angle threshold or the change in slewing angle exceeds the preset single-step slewing angle threshold, the new node is abandoned and random sampling is performed again until an expanded node is obtained that satisfies the requirement that the change in luffing angle does not exceed the preset single-step luffing angle threshold and the change in slewing angle does not exceed the preset single-step slewing angle threshold. The expanded node is then added to the path node sequence. During the random tree search process, at preset intervals of the number of expansion nodes, the straight-line connectivity of the path segments between adjacent nodes in the current path node sequence is detected. If a straight-line path segment between adjacent nodes is detected to pass through any obstacle node, an intermediate node is inserted between the two ends of the straight-line path segment. The position of the intermediate node is selected as the free space position on the straight-line path segment that is farthest from the nearest obstacle node. When the random tree expands to the preset target neighborhood range where the target node is located, the initial hoisting path node sequence from the starting node to the target node is generated backtracking. The straight line segments between adjacent nodes in the initial hoisting path node sequence are taken as path segments. On each path segment, equidistant sampling is performed with a preset discretization step size to generate several discrete path points. The preset discretization step size is less than the second detection step size and not greater than the side length of the smallest leaf node in the octree spatial structure. This ensures that the spatial distance between any two adjacent discrete path points does not exceed the side length of the smallest leaf node, thereby avoiding the omission of local obstacle density changes due to adjacent discrete points crossing multiple octree nodes during the collision detection step size adjustment process.

[0013] Preferably, after inserting an intermediate node between the two endpoints of the straight path segment and before adding the extended node to the path node sequence, the following verification step is included, wherein the node to be verified is either the inserted intermediate node or the extended node to be added: Obtain the previous node of the node to be verified, and calculate the change in luffing angle and slewing angle of the lifting equipment boom corresponding to the movement from the previous node to the node to be verified; when the change in luffing angle exceeds a preset single-step luffing angle threshold or the change in slewing angle exceeds a preset single-step slewing angle threshold, discard the node to be verified; for intermediate nodes, select the second farthest free space position from the nearest obstacle node on the straight path segment as a candidate intermediate node, and repeat the verification of the change in luffing angle and slewing angle. Obtain the fixed station coordinates of the hoisting equipment in the construction site coordinate system and the maximum working radius of the boom in the current hoisting state; Calculate the horizontal distance between the node to be verified and the fixed station coordinates of the hoisting equipment. If the horizontal distance exceeds the maximum working radius, the node to be verified is abandoned. For intermediate nodes, other free space positions on the straight path segment are selected as candidate intermediate nodes in order of distance from the nearest obstacle node. The horizontal distance verification is repeated until an intermediate node is found that satisfies the requirement that the horizontal distance does not exceed the maximum working radius. When it is impossible to select an intermediate node on the straight path segment that satisfies the condition that the horizontal distance does not exceed the maximum working radius, abandon the operation of inserting an intermediate node between the two ends of the straight path segment, and return to the improved fast expanding random tree algorithm to re-expand the previous expanded node in the current path node sequence to generate a new branch path node. Once the node to be verified is obtained that satisfies the requirement that the horizontal distance does not exceed the maximum working radius, the maximum swing offset of the mixing plant module to be lifted at the node to be verified is calculated based on the spatial position of the node to be verified, the geometric dimension parameters of the mixing plant module to be lifted, and the current lifting posture. Determine whether the maximum swing offset causes the bounding box of the mixing plant module to be hoisted to intersect with the bounding boxes of obstacles in the node to be verified and its adjacent nodes in the octree spatial structure. If an intersection occurs, the spatial position of the node to be verified is fine-tuned in the direction away from the nearest obstacle. The fine-tuning distance is an integer multiple of the first detection step size (0.05m) until the maximum swing offset no longer causes the bounding boxes to intersect. For intermediate nodes, the fine-tuned intermediate nodes are added to the path node sequence as the final inserted intermediate nodes. For extended nodes, the fine-tuned extended nodes are added to the path node sequence.

[0014] Preferably, after adding the fine-tuned intermediate node as the final inserted intermediate node to the path node sequence, the method further includes: Get the two adjacent path segments formed by the last inserted intermediate node in the current path node sequence and its adjacent previous and next nodes. Calculate the path turning angle of two adjacent path segments at the final inserted intermediate node. When the path turning angle exceeds the preset turning angle threshold, select a first control point on the first path segment between the final inserted intermediate node and the previous node, and select a second control point on the second path segment between the final inserted intermediate node and the next node. Use a cubic Bézier curve to perform a local smooth transition of the two adjacent path segments near the final inserted intermediate node to generate a smooth transition curve segment. The first control point is located on the first path segment and is at a first preset distance from the final inserted intermediate node, and the second control point is located on the second path segment and is at a second preset distance from the final inserted intermediate node. The smooth transition curve segment is discretized using a sampling interval equal to the first detection step size in collision detection to obtain a sequence of transition discrete points. Intersection tests are then performed on the bounding boxes of the mixing plant module to be hoisted, the bounding boxes of the moving parts of the hoisting equipment, and the bounding boxes of obstacles in the corresponding area of ​​the octree spatial structure at each transition discrete point. If a collision is detected at any transition discrete point, the first preset distance and the second preset distance are increased, the smooth transition curve segment is regenerated and discretized and collision detection is performed until a collision-free smooth transition curve segment is obtained; the part corresponding to the smooth transition curve segment in two adjacent path segments is replaced with the collision-free smooth transition curve segment, and the transition discrete point sequence is incorporated into the discrete path point sequence of the initial hoisting path, and the connection relationship between adjacent discrete path points is updated synchronously.

[0015] The present invention has at least the following beneficial effects: First, regarding the significant differences in the geometric dimensions of the mixing plant modules, if a single bounding box is uniformly used in collision detection, slender, irregularly shaped modules are prone to generating false collision alarms in narrow areas due to the loose fit of the bounding boxes. Conversely, tightly bounding boxes result in high computational costs for intersection testing, leading to a decrease in overall planning efficiency. Furthermore, a fixed detection step size struggles to account for the risk of missed detections and redundant calculations in spaces with uneven obstacle density. This invention classifies modules based on the ratio of the maximum to the minimum size of the circumscribed orthogonal cuboid. Different bounding box strategies with varying precision are applied to heterogeneous and regular modules. Furthermore, based on the obstacle distribution density recorded by the octree nodes and the minimum spatial distance within the path segment, the detection step size is dynamically divided into a denser first detection step size and a sparser second detection step size. This allows for fine-grained detection in areas with dense obstacles or tight spacing to prevent missed detections, and coarse-grained detection in sparse, open areas to save computational resources. Thus, while ensuring the reliability of collision detection, the computational cost of overall path planning is effectively reduced.

[0016] Secondly, when using the 3D construction scene model built with BIM software during the design phase for hoisting planning, the actual position and dimensions of the installed structure may differ from the design values ​​due to construction deviations. Furthermore, the dynamic changes of temporary obstacles at the construction site cause discrepancies between the static model and the actual site conditions. This leads to unforeseen collision risks or path unavailability during actual execution of the generated hoisting path. This invention addresses this by deploying a 3D laser scanner before each hoisting operation to acquire site point cloud data and generate a 3D point cloud model of the construction site. The actual position coordinates and geometric dimensions of the installed structure and obstacles are extracted from this data. The positions and dimensions of components exceeding the preset tolerance are updated in the 3D construction scene model. Simultaneously, obstacle records are dynamically added or removed from the model based on the presence or absence of obstacles in the point cloud. Furthermore, the octree structure of the hoisting operation space is redefined, and various bounding boxes are regenerated. This ensures that subsequent collision detection and path planning are always based on a scene model synchronized with the actual site conditions, thereby eliminating the potential for inaccurate detection and path failure caused by inconsistencies between the model and the site.

[0017] Third, when registering and fusing point cloud data acquired by multiple 3D laser scanners from different locations, differences in perspective, environmental occlusion, and dynamic changes on site can easily lead to point cloud layering, misalignment, or even holes in key spatial areas such as vertical lifting channels and horizontal turning channels that the hoisting path must traverse. This results in distortion of the obstacle position and size information extracted from the point cloud. This invention utilizes target spheres deployed at the boundaries and overlapping areas of the hoisting operation space, with the coordinates of their centers pre-determined in the construction site coordinate system using a total station. This provides a high-precision spatial transformation reference for the point clouds of each scanner. After obtaining the initial registered point cloud, a key fusion area is defined for the critical hoisting channel range. Overlapping point clouds within this area undergo adaptive weighted fusion processing that preserves local geometric features, while non-key areas undergo efficiency-prioritized uniform weighted fusion processing. Holes in the point cloud within the key areas are repaired using surface fitting, ensuring that the fused point cloud model possesses a complete and accurately positioned geometric representation within the core spatial areas affecting hoisting safety. This provides a reliable data foundation for the subsequent accurate extraction of obstacle information.

[0018] Fourth, when performing adaptive weighted fusion of overlapping point clouds from different scanners within a key fusion area, if the weights are adjusted only based on local geometric differences without further differentiation of the degree of difference, it is easy to lose obstacle edge details due to over-smoothing at abrupt geometric edges, or to generate false geometric fluctuations in flat areas due to insufficient noise suppression, thus affecting the accuracy of obstacle boundaries extracted from the point cloud. This invention calculates the average normal vector and average curvature within the respective neighborhoods of overlapping point cloud points to obtain the angle between the normal vectors and the curvature difference, and uses this to quantify the degree of local geometric feature differences. When the angle exceeds a threshold, a smaller fusion weight is assigned to protect edge features; when the angle is small but the curvature difference is large, a medium weight is assigned; and when both are small, a larger weight is assigned to smooth noise. Thus, the fusion intensity is adaptively adjusted according to changes in geometric features during the fusion process, ensuring that the fusion result retains key obstacle edge information while effectively suppressing noise interference in flat areas, creating conditions for the accurate extraction of subsequent obstacle geometric boundaries.

[0019] Fifth, when fusing overlapping point clouds in non-critical fusion areas, if the differences in the original point cloud density of different scanners and the differences in the ranging accuracy of each scanner are ignored, the noise in the point cloud acquired by low-density or low-precision scanners can easily contaminate the high-density, high-precision point cloud data, leading to a decrease in the consistency of the geometric representation quality of the fused point cloud model in non-critical areas. This invention calculates the average distance between points in the neighborhood of two overlapping point clouds as a local point cloud density index, and combines this with the nominal ranging accuracy of each scanner. This makes the fusion weight negatively correlated with the local point cloud density and positively correlated with the nominal ranging accuracy of the scanner. This allows the point cloud points contributed by scanners with higher point cloud density and better ranging accuracy to dominate the fusion result, while the weight of point clouds from scanners with sparse density and lower accuracy is correspondingly reduced. Thus, while ensuring the fusion calculation efficiency in non-critical areas, the overall geometric quality and data consistency of the fused point cloud are maintained, providing a foundation for the stable extraction of obstacle information from the point cloud.

[0020] Sixth, when generating the initial hoisting path from the lifting point to the target installation position based on the octree spatial structure, if only a simple spatial straight line connection is used or the path search method does not consider the actual motion constraints of the hoisting equipment, the generated initial path often includes large amplitude changes or rotations that the hoisting equipment boom cannot perform, or there are a large number of infeasible path segments that cross densely obstacle areas. This leads to repeated iterations and adjustments of node positions in the subsequent collision detection and path optimization stages, significantly increasing the overall planning time. This invention, through an improved fast expanding random tree search process, performs constraint verification on the changes in boom luffing angle and slewing angle of each randomly sampled new node, discarding candidate nodes that exceed a preset single-step action threshold. Furthermore, it periodically checks the linear connectivity of adjacent nodes during the search process, inserting the free space position furthest from the nearest obstacle into the straight-line segment crossing the obstacle as an intermediate obstacle-avoidance node. Simultaneously, it ensures that the discretization sampling step size of the final generated path is neither smaller than the sparser step size in collision detection nor larger than the side length of the smallest leaf node in the octree. This makes the initial path node sequence more closely match the motion capability and free space distribution of the hoisting equipment, providing a more feasible initial path for subsequent refined collision detection and effectively reducing the number of subsequent iterative adjustments.

[0021] Seventh, when inserting intermediate obstacle avoidance nodes between the two end nodes during the initial path search phase, if only the free position farthest from the obstacle in the static space is used as the selection criterion, without checking whether the position is within the working radius of the lifting equipment's boom, and without considering the swing offset caused by acceleration and deceleration during the movement of the lifting module, the generated path nodes may be statically collision-free, but in actual lifting operations, they may be unable to reach the target due to exceeding the reach of the boom, or dynamic interference may occur due to the intersection of the module's dynamic swing envelope space with surrounding obstacles. This invention verifies the changes in boom luffing and slewing angles, the horizontal distance between the crane and the fixed position of the hoisting equipment, and the maximum working radius for each intermediate or extended node to be verified. It also estimates the maximum swing offset caused by acceleration and deceleration based on the module's geometry and hoisting posture, and checks whether its dynamic envelope space intersects with obstacles in adjacent octree nodes. For candidate nodes that do not meet the requirements, the positions are reselected or fine-tuned along the direction away from the obstacles until all constraints are met. The resulting path node sequence not only meets the static reachability conditions of the hoisting equipment but also reserves a safety margin to cope with dynamic motion effects, significantly improving the physical feasibility of the initial path.

[0022] Eighth, after fine-tuning the intermediate node to meet the safety margin of dynamic swing, the fine-tuning operation may cause the node to form a large turning angle with the adjacent path segments before and after it. When the hoisting equipment passes through this point, it needs to change the direction of movement drastically, which will cause the crane arm to accelerate and decelerate violently and the hoisting module to swing significantly. This not only increases the difficulty of on-site operation and safety risks, but may also exceed the motion control accuracy range of the equipment due to the discontinuity of the path curvature. This invention adds finely tuned intermediate nodes to the path sequence, calculates the turning angle between adjacent path segments at that node, and when the turning angle exceeds a preset threshold, selects control points on the preceding and following path segments and uses cubic Bézier curves to locally smooth the transition area. It then discretizes the smooth curve segment with a sampling interval equivalent to a close step size for collision detection. If a collision is detected, the distance between control points is dynamically adjusted to regenerate the curve until a collision-free smooth transition curve segment is obtained. This curve segment replaces the corresponding part in the original path, thus ensuring a continuous and smooth geometric curvature change in the final hoisting path while guaranteeing no collisions. This effectively reduces motion impact and module sway amplitude during hoisting, providing a more kinematically friendly path trajectory for actual hoisting operations.

[0023] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Detailed Implementation

[0024] The present invention will now be described in further detail so that those skilled in the art can implement it based on the description.

[0025] In existing modular hoisting path planning technologies for mixing plants, a single bounding box is typically used to geometrically approximate the hoisting modules, such as uniformly using axis-aligned bounding boxes or uniformly using oriented bounding boxes, while a fixed step size is used to perform collision detection on the discretized path. However, the geometric dimensions of the modules in a mixing plant vary significantly. The axis-aligned bounding boxes of slender, irregularly shaped modules contain a large amount of blank space, which can easily generate false collision alarms when performing collision detection in narrow passage areas, forcing the path planning algorithm to detour or make repeated adjustments, increasing the computational cost of path search. While oriented bounding boxes can fit the module geometry more closely, their intersection test computation is significantly higher than that of axis-aligned bounding boxes. If oriented bounding boxes are used indiscriminately for all modules, a considerable computational burden will accumulate throughout the entire path detection process. In addition, the distribution of obstacles in the hoisting operation space at the construction site is uneven. If the fixed step length is set too large, in areas with dense obstacles or in path sections where the distance between the module and the obstacle is close, the critical collision position may fall between two sampling points and be missed, resulting in the risk of missed detection. If the step length is set too small, a large number of redundant detection points will be generated in wide areas with sparse obstacles, resulting in unnecessary intersection test calculations and prolonging the path planning time.

[0026] To address the aforementioned issues, this specific implementation provides a BIM-based modular hoisting path planning and collision detection system for batching plants. First, a 3D construction scene model is constructed using BIM software, including the batching plant module to be hoisted, hoisting equipment, installed structures, and obstacles on the construction site. The geometric dimensions (length, width, and height) of the circumscribed orthogonal cuboid of the batching plant module to be hoisted are obtained from this 3D construction scene model. The ratio of the maximum to the minimum dimension is calculated. When this ratio exceeds a preset threshold of 3:1, the batching plant module to be hoisted is determined to be a heterogeneous module; when the ratio does not exceed the preset threshold, it is determined to be a regular module. Based on the 3D construction scene model, the reachable space from the hoisting point to the target installation position of the batching plant module to be hoisted is defined as the hoisting operation space. An optimized octree algorithm is used to partition and encode this space, generating an octree spatial structure. This optimized octree algorithm, based on the density of obstacles within the hoisting operation space, performs deep subdivision for areas where the obstacle density exceeds a preset depth subdivision threshold, and shallow subdivision or no subdivision for areas where the obstacle density does not exceed the preset depth subdivision threshold. During the subdivision process, the obstacle density value within the space controlled by each octree node is recorded. Obstacles include installed structures, construction site obstacles, and fixed structures of the hoisting equipment. For heterogeneous modules, a hybrid bounding box combining directional bounding boxes and axis-aligned bounding boxes is used for geometric wrapping. For example, axis-aligned bounding boxes are used for the main body of the module, while directional bounding boxes are used to supplement the wrapping of protruding slender parts. For regular modules, axis-aligned bounding boxes are used for geometric wrapping. Simultaneously, all obstacles within the octree spatial structure and the moving parts of the hoisting equipment are geometrically wrapped using axis-aligned bounding boxes. The bounding box parameters of the moving parts of the hoisting equipment are updated synchronously according to the real-time attitude of the equipment at each detection position along the hoisting path. For example, when the boom of the hoisting equipment luffs or rotates, the bounding box size and orientation of its moving parts are adjusted accordingly. Based on a 3D construction scene model and an octree spatial structure, an initial hoisting path is generated for the batching plant module to be hoisted from the lifting point to the target installation position. This initial hoisting path is then discretized to obtain several discrete path points. During the collision detection and optimization process of the initial hoisting path, for the path segment between two adjacent discrete path points, the minimum spatial distance between the bounding box of the batching plant module to be hoisted and the bounding box of the nearest obstacle within that path segment is calculated based on the octree spatial structure.When the obstacle density of any octree node traversed by the path segment is higher than the preset depth subdivision threshold, or the minimum spatial distance is less than the preset distance threshold of 0.2 meters, dense detection is performed using a first detection step size, such as 0.05 meters. When the obstacle density of all octree nodes traversed by the path segment does not exceed the preset depth subdivision threshold, and the minimum spatial distance is not less than the preset distance threshold, sparse detection is performed using a second detection step size, such as 0.2 meters. The first detection step size is smaller than the second detection step size and not greater than the side length of the smallest leaf node in the octree spatial structure, ensuring that the spatial distance between any two adjacent discrete path points does not exceed the side length of the smallest leaf node. According to the adjusted collision detection step size, at the corresponding detection position in each path segment, the bounding box parameters of the moving parts of the hoisting equipment corresponding to the detection position are first updated according to the real-time attitude of the hoisting equipment. Then, the bounding box of the hoisting mixing station module, the bounding box of the hoisting equipment's moving parts, and the obstacle bounding boxes in the corresponding area of ​​the octree spatial structure are simultaneously intersected to determine whether a collision exists. If a collision is detected, the discrete path point at the corresponding location is adjusted. For example, the path point is moved a fine distance away from the colliding obstacle, and the collision detection step size adjustment and intersection test are re-executed on the adjusted path segment until a collision-free hoisting path is obtained. By adaptively selecting the bounding box accuracy based on the module geometry ratio and dynamically adjusting the collision detection step size based on the obstacle density of the octree nodes and the minimum spatial distance within the path segment, this system uses a fine-grained step size for dense detection in areas with dense obstacles or tight spacing to avoid missed detections, and a coarse-grained step size for fast detection in sparse and open areas to reduce redundant calculations. At the same time, hybrid bounding boxes are used to more tightly wrap heterogeneous modules to reduce false collision alarms. Thus, while ensuring the reliability of collision detection, the overall computational cost is effectively controlled, and the conflict between detection accuracy and computational efficiency caused by the large differences in module geometry and uneven distribution of obstacles during the modular hoisting of the mixing plant is resolved.

[0027] In BIM-based hoisting path planning technology, static 3D construction scene models established during the design phase are typically used directly for collision detection and path searching. The positions and dimensions of installed structures in this static model are derived from the design drawings and do not correct for deviations from actual construction conditions. Furthermore, temporary obstacles such as building materials and equipment temporarily stored on the construction site dynamically appear or disappear, and the static model cannot reflect these changes. When the hoisting path planned based on this static model is put into actual operation, discrepancies between the model and the actual site conditions frequently lead to module collisions due to deviations between the actual positions of installed structures and the model, unnecessary detours caused by obstacles recorded in the model no longer existing, and even unforeseen collisions due to newly appearing obstacles missing from the model.

[0028] Continuing with the aforementioned specific implementation methods, this invention, based on the construction of a 3D construction scene model using BIM software that includes the batching plant modules to be hoisted, hoisting equipment, installed structures, and obstacles on the construction site, further performs the following dynamic update steps before the hoisting operation of each batching plant module. First, at least one 3D laser scanner is deployed at the batching plant construction site, and each 3D laser scanner collects point cloud data of the surrounding environment at different locations on the construction site. The point cloud data collected by each 3D laser scanner is registered and fused to generate a 3D point cloud model of the construction site covering the entire hoisting operation space. The actual position coordinates and actual geometric dimensions of the installed structures, as well as the actual position coordinates and actual geometric dimensions of obstacles on the construction site, are extracted from this 3D point cloud model of the construction site. The extraction method can employ point cloud segmentation and feature fitting algorithms; for example, the position and dimensions of structural components such as walls and columns are obtained through planar fitting, and the circumscribed cuboid parameters of stacked materials are obtained through convex hull extraction.

[0029] The extracted actual position coordinates and actual geometric dimensions of the installed structure are compared item by item with the design position coordinates and design geometric dimensions of the corresponding components in the pre-built BIM design model of the installed structure based on the design drawings. When the Euclidean distance between the actual position coordinates and the design position coordinates of a component exceeds a preset position tolerance threshold, such as 5 cm, the position coordinates of the component in the 3D construction scene model are updated to the actual position coordinates. When the difference between the actual geometric dimensions and the design geometric dimensions of a component exceeds a preset dimension tolerance threshold, such as 2 cm, the geometric dimensions of the component in the 3D construction scene model are updated to the actual geometric dimensions. For obstacles on the construction site, the extracted actual position coordinates and actual geometric dimensions are compared with the position coordinates and geometric dimensions of the obstacles already recorded in the 3D construction scene model. If an obstacle is detected in the 3D point cloud model at a certain spatial location, but there is no record of an obstacle at the corresponding location in the current 3D construction scene model, the obstacle is added to the 3D construction scene model. Conversely, if a certain obstacle is recorded in the 3D construction scene model, but no obstacle point cloud is detected at the corresponding position in the 3D point cloud model of the construction site, then it is determined that the obstacle has been removed and deleted from the 3D construction scene model.

[0030] After updating the position and dimensions of the installed structures and adding and removing obstacles at the construction site, the steps of delineating the hoisting operation space based on the updated 3D construction scene model and using an optimized octree algorithm to spatially divide and encode the hoisting operation space are repeated to generate an updated octree spatial structure. Simultaneously, based on the updated 3D construction scene model, the steps of determining the type of the mixing plant module to be hoisted, generating the bounding box of the mixing plant module to be hoisted, and generating bounding boxes for all obstacles, fixed structures, and moving parts of the hoisting equipment within the octree spatial structure are repeated. Through this series of dynamic updates, the position and dimensions of the installed structures in the 3D construction scene model are consistent with the actual construction results, and the presence or absence of temporary obstacles on site is completely synchronized with the actual situation. Subsequent octree spatial structures and various bounding boxes are all generated based on spatial data reflecting the actual site conditions. Therefore, the problem of model-site inconsistency in collision detection and path planning based on a static model is fundamentally solved. The hoisting path no longer causes unexpected collisions or path inaccessibility due to model deviations in actual operations, significantly improving the safety of hoisting operations and the executability of path planning results.

[0031] During the process of collecting construction site data by multiple 3D laser scanners, due to differences in the perspectives of each device, environmental occlusion, and dynamic changes on site, key areas such as the vertical lifting channel and the horizontal turning channel that the hoisting path must pass through are prone to point cloud layering, misalignment, or even hole problems. The extracted obstacle information has deviations, which will cause the updated 3D construction scene model to be distorted and affect the accuracy of subsequent collision detection and path planning.

[0032] Multiple target spheres are deployed at the boundary of the hoisting operation space on the construction site and within the overlapping area of ​​the scanning range of each 3D laser scanner. The coordinates of the center of each target sphere in the construction site coordinate system are pre-determined using a total station. From the point cloud data acquired by each scanner, the spherical point cloud region corresponding to each target sphere is identified and extracted. For each pair of scanners with overlapping scanning areas, based on the spherical point cloud regions of at least three non-collinear target spheres, a least-squares spherical fitting algorithm is used to calculate the coordinates of the center of each target sphere in the respective coordinate systems of the two scanners. These coordinates are then combined with the pre-determined coordinates of the construction site coordinate system to construct a spatial transformation matrix between the two scanners. Using this spatial transformation matrix, point-by-point coordinate transformation is performed on the point cloud data acquired by all scanners, uniformly converting all point cloud data to the construction site coordinate system to obtain the initial registered point cloud. In the initial registration point cloud, a key fusion region is defined for the vertical lifting channel and horizontal turning channel that are essential for the hoisting operation. This region extends vertically from a preset height above the plane of the lifting point to a preset height above the plane of the target installation position. Horizontally, it covers the shortest straight line connecting the lifting point and the target installation position, and extends to both sides of the line by a preset width. Within the defined key fusion region, an adaptive weighted fusion algorithm based on the angle between normal vectors and curvature variations is used to fuse overlapping point clouds from different scanners. In non-key areas outside the key fusion region, a uniform weighted fusion algorithm based on the average point spacing is used to process overlapping point clouds. After all fusion processing is complete, point cloud holes within the key fusion region are filled using a point cloud hole repair algorithm based on a moving least squares surface.

[0033] This implementation process provides a high-precision spatial reference for multi-site cloud registration. It performs refined fusion processing on core areas that affect hoisting safety, while also taking into account the processing efficiency of non-core areas. The resulting point cloud model has a complete geometric representation and accurate spatial positional relationships within the critical channels that hoisting must pass through. This provides reliable data support for the subsequent accurate extraction of installed structure and obstacle information from the point cloud model, avoiding the problem of scene model distortion caused by point cloud data quality defects from the source, and ensuring the accuracy of subsequent construction scene model updates and path planning.

[0034] When performing overlapping point cloud fusion processing in key fusion areas, if the fusion weight is adjusted only based on the angle between the normal vectors and the curvature changes without further distinguishing the degree of local geometric differences, the obstacle edge details are easily lost due to excessive smoothing at the edges of geometric abrupt changes. In flat areas, false geometric fluctuations are also generated due to insufficient noise suppression, which directly affects the accuracy of obstacle boundary extraction.

[0035] Within the key fusion area, for cases where overlapping point cloud points from two different 3D laser scanners exist at the same spatial location, neighborhood spheres are constructed with preset neighborhood radii centered on each of the two point cloud points. The average normal vector and average curvature value of the local point cloud surfaces formed by all point cloud points within each neighborhood sphere are calculated. The angle between the two average normal vectors and the absolute value of the difference between the two average curvature values ​​are then calculated. Based on the calculated angle and absolute value of the curvature difference, fusion weight coefficients are determined hierarchically for the two point cloud points. When the angle between the normal vectors exceeds a preset angle threshold, a first set of weight values ​​is assigned to the two point cloud points. When the angle between the normal vectors does not exceed the preset angle threshold and the absolute value of the curvature difference exceeds a preset curvature difference threshold, a second set of weight values ​​is assigned. When neither value exceeds the corresponding threshold, a third set of weight values ​​is assigned. The smaller weight value among the three sets increases sequentially. Using the determined fusion weight coefficients, the spatial coordinates of the two overlapping point cloud points are weighted and averaged to obtain the fused point cloud spatial coordinates.

[0036] This processing method can adaptively adjust the fusion intensity according to the degree of difference in local geometric features of the point cloud. It retains the detailed information of the original scan in the edge area of ​​abrupt geometric feature changes, avoids the loss of obstacle edge features, and effectively suppresses scanning noise and eliminates false geometric fluctuations in flat and continuous areas. This provides a stable and reliable data foundation for the subsequent accurate extraction of the geometric boundaries of obstacles from the point cloud model, ensuring the accuracy of subsequent construction scene model updates and hoisting path planning.

[0037] After completing the fine fusion of point clouds in key fusion areas, existing technologies often use a simple uniform weighted fusion method for processing overlapping point clouds in non-key fusion areas. This method does not take into account the differences in point cloud density and ranging accuracy between different scanners, which can easily lead to low-density and low-precision point cloud noise polluting high-precision data. This results in a decrease in the geometric quality of the fused point cloud and affects the consistency of obstacle information extraction.

[0038] In non-key fusion areas outside the key fusion region, for cases where overlapping point cloud points from two different 3D laser scanners exist at the same spatial location, search spheres are constructed with the two point cloud points as centers and a preset search radius. The average distance between all point cloud points within each search sphere is calculated and used as the local point cloud density index for the corresponding point cloud point. The ratio of the two local point cloud density indices is then calculated to obtain the point cloud density ratio. Simultaneously, the reciprocal of the ratio of the nominal ranging accuracies of the corresponding scanners for the two point cloud points is used as the scanner accuracy weighting factor. Combining the point cloud density ratio and the scanner accuracy weighting factor, fusion weight coefficients for the two overlapping point cloud points are determined. The fusion weight coefficients are negatively correlated with the local point cloud density index of the corresponding point cloud and positively correlated with the nominal ranging accuracies of the corresponding scanner, and the sum of the two fusion weight coefficients is one. Using the two determined fusion weight coefficients, a weighted average is calculated on the spatial coordinates of the two overlapping point cloud points to obtain the final fused point cloud point spatial coordinates.

[0039] This processing method ensures efficient fusion processing in non-key fusion areas while allowing scanner data with higher point cloud density and better ranging accuracy to dominate the fusion results. It effectively avoids the pollution of the overall data by low-density and low-precision point clouds, maintains the overall geometric quality and data consistency of point clouds in non-key areas, and provides a reliable foundation for the subsequent stable extraction of obstacle information from the point cloud model, ensuring the accuracy and stability of dynamic updates of the construction scene model.

[0040] After completing the registration and fusion of point cloud data at the construction site and the dynamic update of the 3D construction scene model, existing technologies often use simple straight-line connections or search algorithms that do not consider the motion constraints of the hoisting equipment when generating the initial hoisting path. The generated path often includes large-amplitude turning movements that the equipment cannot perform, or infeasible sections that cross densely obstacle areas, which leads to repeated iterations for subsequent optimization and increases the time consumption of path planning.

[0041] In the constructed octree spatial structure, all nodes intersecting with the bounding box of obstacles are first identified and marked, and unmarked nodes are designated as free space nodes. Using the free space node where the lifting point is located as the starting node and the free space node where the target installation location is located as the target node, an improved fast expanding random tree algorithm is used to search for an initial path in free space. Each time a new node is randomly sampled and generated, the algorithm simultaneously calculates the changes in boom amplitude and slewing angle corresponding to the expansion from the current node to the new node. If any change exceeds a preset single-step threshold, the node is abandoned and resampled until an expanded node meeting the threshold requirement is obtained and added to the path node sequence. During the random tree search, every preset number of expanded nodes, the path segments of adjacent nodes in the current path are checked for straight-line connectivity. If a straight path is detected to pass through an obstacle node, an intermediate node is inserted between the two endpoints. The intermediate node is selected as the free space position on the path segment farthest from the nearest obstacle node. When the random tree expands to the preset neighborhood range of the target node, the initial path node sequence is generated by backtracking. Then, the straight line segments between adjacent nodes are taken as independent path segments and discrete path points are generated by equidistant sampling with a preset discretization step size. The discretization step size used is smaller than the coarse step size of the collision detection and not greater than the side length of the smallest leaf node of the octree, ensuring that the distance between adjacent discrete path points does not exceed the side length of the smallest leaf node.

[0042] The initial path generated by this implementation method is fully adapted to the actual movement capability of the hoisting equipment, avoiding infeasible path segments that exceed the equipment's movement threshold from the source. At the same time, through connectivity detection and intermediate node insertion, the initial path avoids areas with dense obstacles in advance, greatly reducing the number of iterations for subsequent collision detection and path optimization. The setting of discretized step size also avoids missing local obstacle density changes in subsequent detection, effectively improving the overall efficiency and reliability of path planning.

[0043] When inserting intermediate obstacle avoidance nodes during the initial path search process, existing technologies only use the free position furthest from the obstacle in the static space as the selection criterion, without checking whether the node is within the working radius of the lifting equipment's boom, and without considering the swing offset during the movement of the lifting module. This can easily lead to a path that is statically collision-free but actually unreachable, or even cause dynamic interference during the lifting process.

[0044] For inserted intermediate nodes and expansion nodes to be added, motion amplitude verification is first performed. The previous node of the node to be verified is obtained, and the changes in the boom luffing angle and slewing angle corresponding to the movement from that node to the node to be verified are calculated. If either change exceeds the preset single-step threshold, the node to be verified is directly abandoned. For intermediate nodes, the second farthest free space position from the nearest obstacle node is selected as a candidate on the corresponding path segment, and the motion amplitude verification is repeated. After completing the motion amplitude verification, the fixed station coordinates of the hoisting equipment in the construction site coordinate system and the maximum working radius of the boom in the current hoisting state are obtained. The horizontal distance between the node to be verified and the fixed station coordinates is calculated. If the horizontal distance exceeds the maximum working radius, the node is abandoned. For intermediate nodes, other free space positions are selected as candidates in order of distance from the nearest obstacle node, and the verification is repeated. If a node that meets the requirements cannot be selected, the intermediate node insertion operation is abandoned, and the random tree search is returned to re-expand the previous expansion node to generate a new branch path node. For nodes that have passed the first two checks, the maximum swing offset of the module caused by the boom's luffing or slewing acceleration / deceleration is calculated, taking into account its spatial position, the geometry of the module to be hoisted, and the current hoisting posture. It is then determined whether this offset will cause the module's bounding box to intersect with the bounding boxes of the corresponding node and the obstacles in the adjacent nodes. If there is a risk of intersection, the node to be checked is finely adjusted in the direction away from the nearest obstacle. The fine-tuning distance is an integer multiple of the collision detection step size, until the swing offset will not cause the bounding boxes to intersect. Finally, nodes that meet all the requirements are added to the path node sequence.

[0045] This implementation process comprehensively verifies the feasibility of path nodes from three dimensions: equipment mobility, operational reach, and hoisting dynamic effects. It ensures that the generated path nodes not only avoid collision risks in static space but also fully adapt to the actual operational capabilities of the hoisting equipment. At the same time, it reserves a safety margin to cope with the dynamic swing of the modules, which greatly improves the physical feasibility of the initial path and effectively reduces the number of replannings caused by path infeasibility. It provides a stable and reliable initial path foundation for subsequent fine collision detection and path optimization.

[0046] After fine-tuning the dynamic swing safety margin of the intermediate node, the fine-tuning operation is prone to causing the node to form an excessive turning angle with the preceding and following path segments. Existing technology has not made targeted smoothing treatment for such path turns. When the hoisting equipment passes through, it needs to change the direction of movement drastically, causing the crane arm to accelerate and decelerate violently and the module to swing significantly. This not only increases the difficulty of on-site operation, but also easily exceeds the motion control accuracy range of the equipment.

[0047] After adding the intermediate nodes that have completed all verifications and fine-tuning to the path node sequence, first obtain the two adjacent path segments formed by the intermediate node and its adjacent preceding and following nodes, and calculate the turning angle of the two paths at the intermediate node. When the turning angle exceeds the preset turning angle threshold, select a first control point on the first path segment between the intermediate node and the preceding node, and select a second control point on the second path segment between the intermediate node and the following node. The two control points are located on their respective path segments and maintain a preset distance from the intermediate node. Use a cubic Bézier curve to perform a local smooth transition in the turning area near the intermediate node, generating a smooth transition curve segment. Then, use a sampling interval equal to the first detection step size of the collision detection to discretize the smooth transition curve segment to obtain a sequence of transition discrete points. Simultaneously, perform intersection tests on the bounding boxes of the mixing plant module to be hoisted, the moving parts of the hoisting equipment, and the obstacles in the corresponding area of ​​the octree spatial structure at each transition discrete point. If a collision is detected at any transition discrete point, the preset distance between the two control points and the intermediate node is increased, a smooth transition curve segment is regenerated, and the discretization and collision detection process is repeated until a smooth transition curve segment without collisions is obtained. Finally, the corresponding turning area in the original path is replaced with the collision-free curve segment, and the sequence of transition discrete points is incorporated into the sequence of discrete path points of the initial hoisting path. The connection relationship between adjacent discrete path points is updated synchronously.

[0048] This process eliminates sharp turns in the path without introducing new collision risks, keeping the curvature changes of the hoisting path continuous and smooth. It effectively reduces the motion impact and module sway amplitude during hoisting operations, significantly reducing the difficulty and safety risks of on-site hoisting operations. It also ensures that the final path trajectory fully matches the motion control precision requirements of the hoisting equipment, providing a more kinematically friendly feasible path for actual on-site hoisting operations.

[0049] Engineering Application Example of a BIM-Based Modular Lifting Path Planning and Collision Detection System for Mixing Plants I. Application Project Overview This application example is implemented in a city with an annual production capacity of 1.8 million cubic meters. 3 The modular construction project for commercial concrete batching plants involves the construction of two HZS180 fully automated commercial concrete production lines. The project adopts a fully modular prefabrication construction mode, dividing the batching plant into 22 prefabricated modules in 5 major categories: mixing host module, batching machine module, powder tank module, control room module, and screw conveyor module. After prefabrication, the modules are transported to the site for hoisting and assembly.

[0050] The core construction challenges and constraints of the project are as follows: 1. Significant differences in module geometry: The powder tank module has a slender structure with dimensions of Φ3.2m×19.2m, and the ratio of the maximum to the minimum size is 6:1; the screw conveyor module has dimensions of 0.8m×0.6m×3.6m, and the size ratio is 4.5:1, both of which are irregularly shaped and heterogeneous modules; the mixing host module and the control room module have an approximately cubic structure with a length-width-height ratio of no more than 2:1, which are regular modules, and a single bounding box strategy cannot simultaneously ensure detection accuracy and efficiency.

[0051] 2. Limited hoisting space: The available hoisting area at the construction site is only 28m wide. It is surrounded by a three-story office building and a fixed fence for raw material storage. The distribution of obstacles such as the existing foundation structure and temporary building material storage area is uneven. The minimum passage gap in the narrow area is only 1.2m, and the fixed step collision detection is prone to missed detection or redundant calculation.

[0052] 3. High precision requirements for hoisting operations: The maximum working height of the module hoisting is 32m, the maximum hoisting radius is 24m, the hoisting equipment is a 250t truck crane with a main boom length of 42m, a rated maximum single-step luffing angle of 3° and a maximum single-step slewing angle of 5°, and the requirements for the kinematic feasibility and safety of the path are extremely high.

[0053] II. The specific operating steps and parameter settings are as follows: (I) Construction and Dynamic Updating of BIM 3D Construction Scene Model Using Revit 2024 BIM software, a 3D construction scene model was built based on the project's construction design drawings. The model includes 22 modules to be hoisted, the full structure of a 250t truck crane, the installed foundation structure, surrounding fixed buildings, and temporary obstacles on site. The model's coordinate system is completely consistent with the construction site's coordinate system.

[0054] Two hours before each module hoisting operation, a dynamic model update was performed: Three FARO Focus S350 3D laser scanners were deployed on the east, west, and south sides of the hoisting operation area at the construction site to collect point cloud data of the site environment at different stations. The scanning resolution was set to 1 / 4 standard resolution, with a nominal ranging accuracy of ±1mm. Six target spheres were placed at the boundary of the hoisting operation space and in the overlapping area scanned by the three scanners. The coordinates of the center of each target sphere in the construction site coordinate system were pre-determined using a Topcon MS05AXII total station, with a measurement accuracy of ±0.5mm. After registering and fusing the collected point cloud data, the actual positions and geometric dimensions of the installed structures and on-site obstacles were extracted. A position tolerance threshold of 5cm and a dimension tolerance threshold of 2cm were set. After comparison with the BIM design model, the positions and dimensions of out-of-tolerance components were updated. Temporary obstacles added / removed on-site were simultaneously added / deleted in the model. After the model is updated, the reachable space from the lifting point to the target installation position of the module to be lifted is defined as the lifting operation space. An optimized octree algorithm is used to divide the operation space. The depth subdivision threshold is set to 15% of the obstacle in the unit space. For areas with an obstacle density higher than 15%, a maximum of 8 layers of depth subdivision is performed. For areas with an obstacle density lower than 15%, 3-4 layers of shallow subdivision are used. Each octree node synchronously records the obstacle distribution density value of the space under its jurisdiction and generates the corresponding octree space structure.

[0055] (II) Module Type Determination and Precise Construction of Bounding Boxes From the updated BIM model, the length, width, and height geometric parameters of the circumscribed orthogonal cuboid of each module to be hoisted are extracted. A size ratio threshold of 3:1 is set, and the ratio of the maximum to the minimum size of the module is calculated. The powder tank module and the screw conveyor module both have size ratios exceeding 3:1, thus they are classified as heterogeneous modules. The mixing host module, batching machine module, and control room module all have size ratios not exceeding 2:1, thus they are classified as regular modules. Heterogeneous modules are geometrically enclosed using a hybrid bounding box combining directional bounding boxes and axis-aligned bounding boxes. The main body of the powder tank module is tightly enclosed using directional bounding boxes, while the dust removal device at the top and the connecting flange at the bottom are supplemented with axis-aligned bounding boxes. Regular modules are geometrically enclosed directly using axis-aligned bounding boxes. Simultaneously, all fixed obstacles within the octagonal spatial structure and the fixed outrigger structure of the truck crane are enclosed using axis-aligned bounding boxes. Moving parts such as the truck crane boom and hook are enclosed using axis-aligned bounding boxes whose parameters are updated synchronously according to the real-time posture of the equipment.

[0056] (III) Hierarchical Registration and Fusion Processing of Point Cloud Data For point cloud data collected by multiple scanners, initial registration is performed based on deployed target spheres: For each pair of scanners with overlapping scanning areas, based on the spherical point cloud region of the same target sphere, a least-squares spherical fitting algorithm is used to calculate the coordinates of the target sphere's center in the respective coordinate systems of the two scanners. Combined with the coordinates of the construction site coordinate system pre-determined by the total station, a spatial transformation matrix between the two scanners is constructed. Through coordinate transformation, all point cloud data are uniformly converted to the construction site coordinate system to obtain the initial registered point cloud. For the lifting path of the module to be lifted, the vertical lifting channel and horizontal turning channel are designated as key fusion areas. The vertical range of this area extends from 2m above the plane of the lifting point to 2m above the plane of the target installation position, and the horizontal range covers the shortest straight line connecting the lifting point and the target installation position, extending 3m to each side of the line. Within the key fusion area, a neighborhood radius of 5cm, a normal vector angle threshold of 15°, and a curvature difference threshold of 0.2 mm are set. -1 An adaptive weighted fusion algorithm based on the angle between normal vectors and curvature variations is employed to perform hierarchical weighted fusion of overlapping point clouds. In non-key fusion areas, a search radius of 10cm is set, and a uniform weighted fusion algorithm based on the mean point spacing is used, combined with the scanner's nominal ranging accuracy to determine the fusion weights, thus completing the fusion of the entire area's point clouds. After fusion, point cloud holes in key fusion areas are filled using a point cloud hole repair algorithm based on moving least squares surfaces to ensure the integrity of the point clouds in key areas.

[0057] (iv) Initial hoisting path generation and node feasibility verification In the octree spatial structure, all nodes intersecting with the bounding box of obstacles are identified and marked as obstacle nodes, and the remaining nodes are free space nodes. Starting with the free space node where the module lifting point is located and the free space node where the target installation location is located as the target node, an improved fast expanding random tree algorithm is used to search for an initial path in free space. Each time the algorithm randomly samples and generates a new node, it simultaneously calculates the change in boom luffing angle and slewing angle corresponding to the expansion from the current node to the new node. When the change exceeds the preset single-step luffing angle threshold of 3° and the single-step slewing angle threshold of 5°, the new node is abandoned and resampled until an expanded node that meets the constraints is obtained and added to the path node sequence. During the random tree search, every 5 nodes, the straight-line connectivity between adjacent nodes of the current path is checked. If a straight-line path segment is detected to pass through an obstacle node, an intermediate node is inserted between the two endpoints. The intermediate node is selected as the free space position on the path segment farthest from the nearest obstacle node. Perform full-process verification on the inserted intermediate nodes and the expansion nodes to be added: First, repeatedly verify the changes in amplitude and rotation angle. If they meet the requirements, verify whether the horizontal distance between the node and the fixed position of the truck crane exceeds the maximum working radius of 24m. If it does, reselect candidate nodes in order of distance from the obstacle from farthest to near. After passing the accessibility verification, combine the module geometry and lifting posture to calculate the maximum swing offset of the module at the node caused by the acceleration and deceleration of the crane boom. Set a swing safety margin of 10cm. If the offset causes the module bounding box to intersect with the obstacle bounding box, fine adjust the point position in the direction away from the obstacle. The fine adjustment distance is an integer multiple of 0.05m until the safety requirements are met. Finally, add the nodes that meet all constraints to the path node sequence.

[0058] (v) Path smoothing and adaptive step size collision detection For intermediate nodes that have been validated and added to the path sequence, calculate their turning angles relative to the preceding and following path segments. Set a turning angle threshold of 30°. When the turning angle exceeds the threshold, select control points on the intermediate node and the preceding and following path segments respectively. Use a cubic Bézier curve to perform a local smooth transition in the turning area. Set the initial distance between the control point and the intermediate node to 0.5m. Discretize the smooth transition curve segment using a sampling interval of 0.05m. Perform collision detection on each transition discrete point. If a collision is detected, increase the distance between the control point and the intermediate node, regenerate the smooth curve, and repeat the detection until a collision-free smooth transition curve segment is obtained. Replace the turning area in the original path with this curve segment, and incorporate the transition discrete point into the discrete path point sequence of the initial path.

[0059] For the complete initial hoisting path, equidistant sampling is performed using a discretization step size of 0.1m. This step size is less than the collision detection coarse step size and no greater than the side length of the smallest leaf node in the octree (0.2m), resulting in a complete discrete path point sequence. For path segments between two adjacent discrete path points, the minimum spatial distance between the bounding box of the module and the bounding box of the nearest obstacle within the path segment is calculated based on the octree spatial structure. The collision detection step size is adaptively adjusted: when the obstacle density of any octree node traversed by the path segment is higher than 15%, or the minimum spatial distance is less than 0.2m, a first detection step size of 0.05m is used; when the obstacle density of all octree nodes traversed by the path segment does not exceed 15%, and the minimum spatial distance is not less than 0.2m, a second detection step size of 0.2m is used.

[0060] According to the adjusted collision detection step size, at the corresponding detection position of each path segment, the bounding box parameters of the moving parts of the truck crane are first updated according to the real-time attitude of the truck crane at the detection position. Then, the intersection test of the module bounding box, the bounding box of the moving parts of the truck crane and the bounding box of the obstacles in the corresponding area is performed simultaneously to determine whether there is a collision. If a collision is detected, the discrete path points at the corresponding positions are adjusted, and the step size adjustment and intersection test are re-executed on the adjusted path segment. Finally, the optimal hoisting path with no collisions throughout is obtained and output to the on-site hoisting personnel for execution.

[0061] Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details.

Claims

1. A BIM-based modular hoisting path planning and collision detection system for a mixing plant, characterized in that, include: Based on whether the ratio of the maximum to the minimum size in the geometric parameters of the circumscribed orthogonal cuboid of the mixing plant module to be hoisted exceeds a preset ratio threshold, the module is determined to be a heterogeneous module or a regular module. The hoisting operation space is delineated based on the BIM model. An optimized octree algorithm is used to divide the hoisting operation space into spatial parts, generating an octree spatial structure. The obstacle distribution density value within the space under the jurisdiction of each octree node is recorded. For heterogeneous modules, a hybrid bounding box combining directional bounding boxes and axis-aligned bounding boxes is used for geometric wrapping; for regular modules, axis-aligned bounding boxes are used for geometric wrapping; and for obstacles and moving parts of hoisting equipment, axis-aligned bounding boxes are used for geometric wrapping. Generate an initial hoisting path from the lifting point to the target installation position, and discretize the initial hoisting path to obtain discrete path points; During the collision detection process of the initial hoisting path, for the path segment between two adjacent discrete path points, the minimum spatial distance between the bounding box of the module and the bounding box of the nearest obstacle within the path segment is calculated based on the octree spatial structure. When the obstacle density of any octree node traversed by the path segment is higher than the preset depth subdivision threshold, or the minimum spatial distance is less than the preset distance threshold, a first detection step size is adopted; otherwise, a second detection step size is adopted; the first detection step size is less than the second detection step size. According to the adjusted detection step size, the intersection test of the module bounding box, the bounding box of the hoisting equipment moving parts, and the bounding box of the obstacle is carried out at each detection position. If a collision is detected, the discrete path points at the corresponding positions are adjusted until a hoisting path without collisions is obtained.

2. The BIM-based modular hoisting path planning and collision detection system for mixing plants according to claim 1, characterized in that, Before the hoisting operation of each batching plant module to be hoisted, the following dynamic update steps shall be performed: A 3D laser scanner is deployed at the construction site to collect point cloud data, which is then registered and fused to generate a 3D point cloud model of the construction site covering the hoisting operation space. Extract the actual position coordinates and actual geometric dimensions of the installed structure, as well as the actual position coordinates and actual geometric dimensions of obstacles at the construction site, from the 3D point cloud model; The actual position coordinates and actual geometric dimensions of the installed structure are extracted and compared with the design position coordinates and design geometric dimensions of the corresponding components in the BIM design model. When the position deviation exceeds the preset position tolerance threshold or the size deviation exceeds the preset size tolerance threshold, the position or size of the corresponding component in the three-dimensional construction scene model is updated to the actual data. The actual location coordinates and actual geometric dimensions of the obstacles extracted from the construction site are compared with the obstacle data already recorded in the 3D construction scene model. If the obstacle exists in the point cloud but not in the model, it is added to the model; if the obstacle exists in the model but not in the point cloud, it is removed from the model. Based on the updated 3D construction scene model, the hoisting operation space is redefined and an optimized octree algorithm is used for spatial partitioning and encoding to generate an updated octree spatial structure. The type determination of the mixing plant module to be hoisted, the bounding box generation, and the bounding box generation of all obstacles, fixed structures and moving parts of the hoisting equipment within the octree spatial structure are then re-executed.

3. The BIM-based modular hoisting path planning and collision detection system for mixing plants according to claim 2, characterized in that, The point cloud data collected by each 3D laser scanner is registered and fused to generate a 3D point cloud model of the construction site covering the hoisting operation space, which further includes: Target spheres are set up at the boundaries of the hoisting operation space and in the overlapping area of ​​the scanner. The coordinates of the center of each target sphere in the coordinate system of the construction site are determined in advance using a total station. Based on the spherical point cloud regions of the same target sphere in the point clouds of different scanners, the coordinates of the sphere center are calculated by least squares spherical fitting, and a spatial transformation matrix is ​​constructed by combining the pre-determined construction site coordinates to uniformly transform the point clouds of each scanner to the construction site coordinate system, thus obtaining the initial registration point cloud; In the initial registration point cloud, key fusion areas are defined for the vertical lifting channels and horizontal turning channels that the hoisting path must pass through, and the rest are non-key fusion areas; In key fusion areas, an adaptive weighted fusion algorithm based on the angle between normal vectors and curvature changes is used to fuse overlapping point clouds; in non-key fusion areas, a uniform weighted fusion algorithm based on the average point spacing is used to fuse overlapping point clouds. For point cloud holes in key fusion areas, a point cloud hole repair algorithm based on moving least squares surface is used for filling.

4. The BIM-based modular hoisting path planning and collision detection system for mixing plants according to claim 3, characterized in that, An adaptive weighted fusion algorithm is used to fuse overlapping point clouds within key fusion areas, further including: For the first and second point cloud points that overlap, construct a neighborhood sphere with each point as the center, and calculate the average normal vector and average curvature value of the local point cloud surface within each neighborhood sphere. Calculate the angle between the two average normal vectors and the absolute value of the difference between the two average curvature values; The fusion weights are determined hierarchically based on the included angle and the absolute value of the difference: when the included angle exceeds a preset included angle threshold, a first set of weight values ​​is assigned; when the included angle does not exceed the threshold and the absolute value of the curvature difference exceeds a preset curvature difference threshold, a second set of weight values ​​is assigned; when neither of the two exceeds the corresponding threshold, a third set of weight values ​​is assigned; wherein, the smaller weight value in the first set of weight values ​​is smaller than the smaller weight value in the second set of weight values, and the smaller weight value in the second set of weight values ​​is smaller than the smaller weight value in the third set of weight values; The spatial coordinates of the two point cloud points are weighted and averaged using the determined weighting coefficients to obtain the fused point cloud points.

5. The BIM-based modular hoisting path planning and collision detection system for mixing plants according to claim 3, characterized in that, In non-key fusion regions, a uniform weighted fusion algorithm is used to fuse overlapping point clouds, further including: For the first and second point cloud points that overlap, a search sphere is constructed with each point as the center, and the average distance between the point cloud points in each search sphere is calculated as a local point cloud density index. Obtain the nominal ranging accuracy of the scanner corresponding to each point cloud point; The fusion weight coefficient is determined based on the local point cloud density index and the nominal ranging accuracy. The fusion weight coefficient is negatively correlated with the local point cloud density index and positively correlated with the nominal ranging accuracy. The spatial coordinates of the two point cloud points are weighted and averaged using the determined weighting coefficients to obtain the fused point cloud points.

6. The BIM-based modular hoisting path planning and collision detection system for mixing plants according to claim 1, characterized in that, The initial hoisting path is generated and discretized, and further includes: In an octree spatial structure, nodes that do not intersect with the bounding box of obstacles are marked as free space nodes; Using the free space node where the lifting point is located as the starting node and the free space node where the target installation location is located as the target node, an improved fast expanding random tree algorithm is used to search for a path in free space. Each time a new node is randomly sampled and generated, the change in boom luffing angle and slewing angle corresponding to the expansion from the current node to the new node are calculated. When the change in luffing angle exceeds the preset single-step luffing angle threshold or the change in slewing angle exceeds the preset single-step slewing angle threshold, the node is abandoned and resampled. During the search process, every preset number of expansion nodes, the connectivity of the straight line segments between adjacent nodes in the current path is checked. If the straight line segment passes through an obstacle node, an intermediate node is inserted between the two end nodes. The intermediate node is selected as the free space position on the straight line segment that is farthest from the nearest obstacle node. When the random tree expands to the preset neighborhood range of the target node, the initial hoisting path node sequence is generated by backtracking. The straight line segments between adjacent nodes are used as path segments, and discrete path points are generated by equidistant sampling with a preset discretization step size. The preset discretization step size is less than the second detection step size and not greater than the side length of the smallest leaf node in the octree spatial structure.

7. The BIM-based modular hoisting path planning and collision detection system for mixing plants according to claim 6, characterized in that, Before inserting intermediate nodes or adding extended nodes, the following validation steps are also included for the nodes to be validated: Obtain the previous node of the node to be verified, calculate the change in boom luffing angle and slewing angle corresponding to the movement from the previous node to the node to be verified, and discard the node to be verified and reselect a candidate node when the change in luffing angle exceeds the preset single-step luffing angle threshold or the change in slewing angle exceeds the preset single-step slewing angle threshold. Obtain the coordinates of the fixed station position of the hoisting equipment and the current maximum working radius. Calculate the horizontal distance between the node to be verified and the fixed station position. When the horizontal distance exceeds the maximum working radius, abandon the node to be verified and reselect a candidate node. After the above verification, based on the spatial position of the node to be verified, the module's geometric dimensions, and the current hoisting posture, the maximum swing offset of the module caused by the boom's luffing or slewing acceleration and deceleration is calculated. If the swing offset causes the module's bounding box to intersect with the bounding boxes of obstacles in the node and adjacent nodes, the node to be verified is fine-tuned in the direction away from the nearest obstacle. The fine-tuning distance is an integer multiple of the first detection step size until the requirement of no intersection is met. The fine-tuned node is then added to the path node sequence.

8. The BIM-based modular hoisting path planning and collision detection system for mixing plants according to claim 7, characterized in that, After adding the fine-tuned intermediate nodes to the path node sequence, the following is also included: Obtain the two path segments formed by the intermediate node and its adjacent preceding and following nodes, and calculate the path turning angle at the intermediate node. When the turning angle exceeds the preset turning angle threshold, a first control point is selected on the path segment between the middle node and the previous node, and a second control point is selected on the path segment between the middle node and the next node. A cubic Bézier curve is used to perform a local smooth transition in the turning area to generate a smooth transition curve segment. The smooth transition curve segment is discretized using a sampling interval equal to the first detection step size to obtain a sequence of transition discrete points, and an intersection test is performed on each transition discrete point. If a collision is detected, the distance between the first and second control points and the intermediate node is increased, the smooth transition curve segment is regenerated, and discretization and collision detection are repeated until a collision-free smooth transition curve segment is obtained. Replace the corresponding turning point in the original path with the collision-free smooth transition curve segment, and incorporate the transition discrete point sequence into the discrete path point sequence of the initial hoisting path.