Dynamic obstacle prediction and obstacle avoidance path planning method based on multi-sensor fusion

Abstract

This invention relates to the field of navigation assistance technology, and in particular to a dynamic obstacle prediction and obstacle avoidance path planning method based on multi-sensor fusion. This invention achieves comprehensive perception of dynamic obstacles through multiple sensors, providing a rich and reliable data foundation for subsequent processing. At the same time, it predicts the motion of dynamic obstacles through data processing fusion and model prediction, fully considering the motion inertia of obstacles to make their trajectory prediction closer to real physical laws. Furthermore, by "delineating the final obstacle avoidance target area", the global path search problem is transformed into a local but sufficiently large spatial search problem, significantly reducing the number of grids to be processed and the amount of search computation, meeting the stringent real-time requirements of mobile devices. Moreover, by constructing a cost function, it avoids areas that obstacles may occupy in the future in real time, thereby generating an obstacle avoidance planning path, which helps to improve path safety.

Description

Technical Field

[0001] This invention relates to the field of navigation assistance technology, and in particular to a method for dynamic obstacle prediction and obstacle avoidance path planning based on multi-sensor fusion. Background Technology

[0002] With the development of artificial intelligence technology, smart mobile devices (such as autonomous vehicles, service robots, and drones) are increasingly widely used in industries, transportation, and people's livelihoods. As the core function of smart mobile device navigation, obstacle avoidance path planning directly determines the safety and reliability of the device in complex environments.

[0003] Currently, obstacle avoidance path planning methods mostly rely on a single sensor (such as LiDAR or visual camera) to obtain environmental information. However, a single sensor has obvious limitations. In addition, traditional methods often adopt a "real-time avoidance" mode to handle dynamic obstacles, planning the path only based on the current state of the obstacle without considering the future movement trend of the obstacle. This leads to problems such as path planning lag and untimely obstacle avoidance in scenarios where obstacles move quickly (such as intersections and densely populated areas). Therefore, a solution is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a dynamic obstacle prediction and obstacle avoidance path planning method based on multi-sensor fusion, in order to solve the aforementioned technical defects.

[0005] The objective of this invention can be achieved through the following technical solution: a dynamic obstacle prediction and obstacle avoidance path planning method based on multi-sensor fusion, comprising the following steps:

[0006] Step 1: Acquire dynamic obstacle information, location information, and device motion status information in the environment through multiple sensors;

[0007] Step 2: Preprocess the multi-source sensor data collected in Step 1 to obtain standardized data;

[0008] Step 3: Use a preset fusion model to fuse the standardized data obtained in Step 2 and output the fusion perception result;

[0009] Step 4: Dynamic obstacle motion prediction: Based on the fusion perception results output in Step 3, retrieve the preset prediction model, input the historical motion state sequence of the obstacle, and obtain the motion trajectory information of the dynamic obstacle within the prediction time step T.

[0010] Step 5: Obtain the prediction time step T and the discrete trajectory dataset;

[0011] Step Six: Construction and analysis of static and dynamic restricted areas and acquisition of raster maps showing impassable and passable zones;

[0012] Step 7: Plan the obstacle avoidance path using a cost function to obtain the obstacle avoidance planning path.

[0013] Preferably, the process of obtaining the prediction time step T and the discrete trajectory dataset is as follows:

[0014] Acquire the current motion parameters of dynamic obstacles, and simultaneously read the current mobile device's driving speed v2 and the effective detection distance S of the multi-sensor;

[0015] The predicted time step T is calculated using the formula: (effective detection distance S + preset initial safety buffer distance) / min(v1, v2). If there is a decimal, it is rounded up to the nearest integer, and T is limited to [1s, 5s].

[0016] Further analysis of driving speed v2 and obstacle speed v1 shows that if driving speed v2 is greater than or equal to the preset driving speed threshold or obstacle speed v1 is greater than or equal to the preset obstacle speed threshold, then T∈[1s,6s].

[0017] In the prediction time step T, n discrete time points t are uniformly selected, t = 1, 2, ..., n, where n = T / Δt, and Δt is the preset time interval;

[0018] Output the predicted position coordinates (xt, yt, zt) and motion state change of the dynamic obstacle at each discrete time point t. If the dynamic obstacle is a ground-based dynamic obstacle, then zt = 0, forming a discrete trajectory dataset with time and position correlation.

[0019] Preferably, the construction and analysis process of the static + dynamic restricted area is as follows:

[0020] Obtain the basic attribute data of the dynamic obstacle, which includes obstacle type and size parameters (length L, width W, height H).

[0021] Constructing scene bounding boxes for different devices based on discrete trajectory datasets:

[0022] Dynamic obstacles construct a two-dimensional bounding box for ground equipment, centered at (xt, yt), with boundaries x∈[xt-L / 2, xt+L / 2] and y∈[yt-W / 2, yt+W / 2].

[0023] Dynamic obstacles construct a three-dimensional bounding box for aerial equipment, with an additional z-axis boundary, z∈[zt-H / 2, zt+H / 2].

[0024] Preferably, the final safe buffer distance Da = max (maximum device outline size × preset multiple, maximum travel distance of preset response time).

[0025] Extend the scene bounding box of each discrete time point t outward along all boundaries to a final safe buffer distance Da, forming a dynamic restricted area for that discrete time point t, and add a timestamp label t to the dynamic restricted area (marking the effective time interval [tt', t+t'], at which time the grid is impassable, and t' is the time it takes for the device to pass through a single grid).

[0026] By acquiring the position and outline information of fixed obstacles through multiple sensors, a static restricted area is constructed, and then spatially superimposed with the dynamic restricted area to obtain all static and dynamic restricted areas.

[0027] Preferably, the process for obtaining the impassable and passable raster maps is as follows:

[0028] With the current location q (xq, yq, zq) of the current device as the center, construct a circular / spherical area with a preset radius of R1 as the starting passable range, and with the navigation target point m (xm, ym, zm) as the center, construct a circular / spherical area with a preset radius of R2 as the ending passable range;

[0029] Calculate the external tangent line / external plane of the passable range at the starting point and the passable range at the ending point. Based on the external tangent line / external plane, expand the width Wk = preset R1 + final safety buffer distance Da to both sides to form a closed polygon / polyhedron region, which is the initial boundary of the obstacle avoidance target area.

[0030] Preferably, the initial boundary is checked to see if it contains all prohibited areas. If it does, a normal signal is generated; otherwise, an abnormal signal is generated.

[0031] When a normal signal is generated, the final target area boundary is determined, and the final obstacle avoidance target area is obtained.

[0032] The final obstacle avoidance target area is divided into P×K (two-dimensional) or P×K×V (three-dimensional) grids according to the preset grid size, and a unique index is assigned to each grid.

[0033] Initial marking: Traverse all grids. If the center of a grid falls within a restricted area (combined with the timestamp label t), mark it as impassable and assign the corresponding grid a value of 1.

[0034] Secondary labeling: For unlabeled grids (passable), calculate the minimum distance dmin from its center to all restricted areas. Based on the minimum distance dmin, determine whether a grid is a low-security grid that is passable and assign it a value of 0.5, or a high-security grid that is passable and assign it a value of 0. This will result in a grid map of grids that are both passable and impassable.

[0035] Preferably, the obstacle avoidance planning path acquisition and analysis process is as follows:

[0036] Construct a cost function based on a raster map that is both impassable and passable: ;

[0037] Where: α and β are the preset weight coefficients of JL(o) ​​and QZ(o), respectively; JL(o) ​​represents the Euclidean distance from the starting point q to the current node o; QZ(o) represents the straight-line distance from the current node o to the navigation target point m; DF(q) represents the dynamic risk cost of the current node q.

[0038] Preferably, further analysis is performed based on the cost function:

[0039] SS1: Initialize the open list (stores nodes to be searched) and the closed list (stores nodes that have been searched), add the starting point m to the open list, and calculate the cost function value of the starting point m;

[0040] SS2: Select the node with the smallest cost function value from the open list as the current node o, and move it from the open list to the closed list;

[0041] SS3: Generate neighboring nodes o+1 of the current node o (within the forward direction and the device's allowed turning angle range), and analyze each neighboring node o+1:

[0042] If the adjacent node o+1 is impassable or already in the closed list, skip that node;

[0043] Calculate the cost function value of the neighboring node o+1. If the neighboring node is not in the open list, add it to the open list and record "current node" as its parent node.

[0044] If the neighboring node o+1 is already in the open list, compare the newly calculated cost function value of the neighboring node o+1 with the original cost function value. If the new cost function value is less than the original cost function value, then update the parameter information of the neighboring node o+1.

[0045] Repeat steps SS2-SS3 until the endpoint is added to the closed list. Backtrack the parent node to obtain the obstacle avoidance planning path.

[0046] The beneficial effects of this invention are as follows:

[0047] (1) This invention achieves all-round perception of dynamic obstacles through multiple sensors, providing a rich and reliable data foundation for subsequent processing. At the same time, it performs motion prediction of dynamic obstacles through data processing fusion and model prediction, fully considering the motion inertia (historical state sequence) of obstacles, so that the trajectory prediction is closer to the real physical laws.

[0048] (2) The present invention also transforms the global path search problem into a local but sufficiently large spatial search problem by “defining the final obstacle avoidance target area”, which significantly reduces the number of grids to be processed and the amount of search computation, meets the stringent real-time requirements of mobile devices, and avoids the area that obstacles may occupy in the future in real time by constructing a cost function, thereby generating an obstacle avoidance planning path, which helps to improve path safety. Attached Figure Description

[0049] The invention will now be further described with reference to the accompanying drawings;

[0050] Figure 1 This is a reference diagram of the method of the present invention;

[0051] Figure 2 This is a partial analysis diagram of Embodiment 3 of the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments;

[0054] Example 1: Please refer to Figures 1 to 2 As shown, this invention is a dynamic obstacle prediction and obstacle avoidance path planning method based on multi-sensor fusion, including the following steps:

[0055] Step 1: Acquire dynamic obstacle information, location information, and device motion status information in the environment through multiple sensors;

[0056] Step 2: Preprocess the multi-source sensor data collected in Step 1 to obtain standardized data;

[0057] Step 3: Use a preset fusion model (Federated Kalman filter algorithm) to fuse the standardized data obtained in Step 2 and output the fusion perception result;

[0058] Step 4: Dynamic obstacle motion prediction: Based on the fusion perception results output in Step 3, the preset prediction model (improved long short-term memory network (LSTM) model) is retrieved, and the historical motion state sequence of the obstacle is input (including position (X, Y, Z), velocity (Vx, Vy), acceleration (Ax, Ay), time step, etc.) to obtain the motion trajectory information of the dynamic obstacle within the predicted time step T (including position (Xt, Yt, Zt), velocity (Vxt, Vyt) etc. at each future time).

[0059] Step 5: Obtain the prediction time step T and the discrete trajectory dataset;

[0060] Step Six: Construction and analysis of static and dynamic restricted areas and acquisition of raster maps showing impassable and passable zones;

[0061] Step 7: Plan the obstacle avoidance path using a cost function to obtain the obstacle avoidance planning path.

[0062] In step one: the multiple sensors include at least LiDAR, vision camera, millimeter-wave radar, IMU (inertial measurement unit), etc.

[0063] Among them, lidar is used to acquire three-dimensional point cloud data of obstacles to determine the position and size of dynamic obstacles, vision camera is used to acquire environmental image data to extract semantic information and pixel coordinates of dynamic obstacles, and millimeter-wave radar is used to acquire distance and velocity data of dynamic obstacles to capture the motion state of dynamic targets.

[0064] In step two: preprocessing includes noise filtering, time synchronization, and spatial registration;

[0065] Noise filtering: Voxel filtering is used to remove redundant points from LiDAR point cloud data, reducing data volume while preserving dynamic obstacle contour features; Gaussian filtering is used to eliminate Gaussian noise from visual camera image data, avoiding noise interference with semantic segmentation and feature extraction; median filtering is used to remove impulse noise from millimeter-wave radar data, ensuring the stability of dynamic obstacle distance and velocity data; Kalman filtering is used to suppress drift error from inertial measurement equipment data, improving the measurement accuracy of the equipment's own motion state.

[0066] Time synchronization: Using the timestamp of the inertial measurement equipment as a reference (which has a high sampling frequency and good time stability), linear interpolation is used to time-calibrate the data of LiDAR, vision camera and millimeter-wave radar; for the difference in sampling frequency of different sensors, interpolation calculation is used to complete the data of each sensor at the same time node, to ensure the time alignment of multi-source data and avoid fusion deviation caused by time difference;

[0067] Spatial registration: Based on the preset sensor extrinsic parameter matrix (obtained through offline calibration, including the translation and rotation parameters of each sensor relative to the device coordinate system), the data collected by LiDAR, millimeter-wave radar and vision camera are converted to a unified device coordinate system;

[0068] For example, the pixel coordinates of the vision camera can be converted into three-dimensional coordinates in the device coordinate system through intrinsic and extrinsic parameter matrices, and the polar coordinate data of the millimeter-wave radar can be converted into rectangular coordinate data, thereby realizing spatial alignment of multi-source data and laying the foundation for subsequent data fusion.

[0069] In step three: the fusion perception results include dynamic obstacle position, speed, size, environmental semantic information, etc.

[0070] Example 2: Step 5: The process of obtaining the prediction time step T and the discrete trajectory dataset: Obtain the current motion parameters of the dynamic obstacle (including velocity v1, acceleration a1, and direction angle θ1), and at the same time read the current mobile device's driving speed v2 and the effective detection distance S of the multi-sensor (the minimum value among the detection distances of LiDAR, visual sensors, etc.).

[0071] The predicted time step T is calculated using the formula: (effective detection distance S + preset initial safety buffer distance) / min(v1, v2). If there is a decimal, it is rounded up to the nearest integer, and T is limited to [1s, 5s].

[0072] Further analysis of driving speed v2 and obstacle speed v1 shows that if driving speed v2 is greater than or equal to the preset driving speed threshold or obstacle speed v1 is greater than or equal to the preset obstacle speed threshold, then T∈[1s,6s].

[0073] In the prediction time step T, n discrete time points t are uniformly selected, t = 1, 2, ..., n, where n = T / Δt, and Δt is the preset time interval;

[0074] Output the predicted position coordinates (xt, yt, zt) of the dynamic obstacle at each discrete time point t and the change in motion state (such as velocity increment, direction angle increment, etc.). If the dynamic obstacle is a ground dynamic obstacle, then zt = 0, forming a discrete trajectory dataset with time and position correlation.

[0075] Step Six: Construction and Analysis Process of Static + Dynamic Restricted Zones:

[0076] The basic attribute data of dynamic obstacles is obtained, including obstacle type (such as pedestrians, vehicles, etc.) and size parameters (length L, width W, height H, obtained by LiDAR point cloud fitting or visual sensor image segmentation).

[0077] Constructing scene bounding boxes for different devices based on discrete trajectory datasets:

[0078] Dynamic obstacles construct a two-dimensional bounding box for ground equipment (such as wheeled robots), centered at (xt, yt), with boundaries x∈[xt-L / 2, xt+L / 2] and y∈[yt-W / 2, yt+W / 2].

[0079] Dynamic obstacles construct a three-dimensional bounding box for aerial devices (such as drones), with an additional z-axis boundary, z∈[zt-H / 2, zt+H / 2], to ensure that the two-dimensional or three-dimensional bounding box can completely enclose the obstacle without redundant space;

[0080] Obtain the final safe buffer distance Da = max(maximum device outline size × preset multiple (e.g., 1.2), maximum travel distance based on preset response time (unit: seconds);

[0081] The scene bounding box at each discrete time point t is extended outward along all boundaries by a final safe buffer distance Da, forming a dynamic restricted area at that discrete time point t. A timestamp label t is added to the dynamic restricted area (marking the effective time interval [tt', t+t'], where the grid is impassable and t' is the time it takes for the device to pass through a single grid), to avoid misjudging restricted areas with non-overlapping times as current obstacles.

[0082] By acquiring the position and outline information of fixed obstacles (such as walls and pillars) through multiple sensors, a static restricted area is constructed and spatially superimposed with the dynamic restricted area to ensure that the target area contains all static and dynamic restricted areas.

[0083] This is accompanied by a rationality analysis of the restricted area: for each discrete time point t, it is determined whether the restricted area covers the starting point or the ending point of the equipment. If it covers the starting point, a safety control signal is generated.

[0084] If the endpoint is covered, a step extension signal is generated;

[0085] In response to the safety control signal, the final safety buffer distance Da is adjusted to ensure that the initial position of the equipment is passable. In response to the step extension signal, the prediction time step T is extended until the restricted area leaves the endpoint range to avoid invalid scenarios of "target point unreachable".

[0086] The process of obtaining raster images for both impassable and passable sections is as follows:

[0087] Centered on the current location q (xq, yq, zq) of the current device, construct a circular (ground) / spherical (air) area with a preset radius of R1 as the starting passable range, and centered on the navigation target point m (xm, ym, zm), construct a circular / spherical (air) area with a preset radius of R2 as the ending passable range;

[0088] Calculate the external tangent (two-dimensional) / external tangent (three-dimensional) of the passable range at the starting point and the passable range at the ending point. Based on the tangent / tangent, expand the width Wk = preset R1 + final safety buffer distance Da to both sides to ensure that the equipment does not exceed the area range when turning, forming a closed polygon (ground) / polyhedron (air) area, which is the initial boundary of the obstacle avoidance target area.

[0089] Check if the initial boundary includes all prohibited areas. If so, generate a normal signal; otherwise, generate an abnormal signal.

[0090] Responding to normal or abnormal signals and displaying the preset warning text corresponding to the normal or abnormal signal, reminding operation and management personnel to rationally manage and correct the obstacle avoidance target area;

[0091] When a normal signal is generated, the final target area boundary is determined, and the final obstacle avoidance target area is obtained.

[0092] The final obstacle avoidance target area is divided into P×K (two-dimensional) or P×K×V (three-dimensional) grids according to the preset grid size, and a unique index is assigned to each grid (e.g., the index of a two-dimensional grid is (i,h), where i is the grid number in the X-axis direction and h is the grid number in the Y-axis direction; if the index of a three-dimensional grid is (i,h,b), b is the grid number in the Z-axis direction).

[0093] Initial marking: Traverse all grids. If the center of a grid falls within a restricted area (combined with the timestamp label t), mark it as impassable and assign the corresponding grid a value of 1.

[0094] Secondary marking: For unmarked grids (passable), calculate the minimum distance dmin from its center to all prohibited areas. If the minimum distance dmin is less than the preset minimum distance threshold, it is marked as a low-security grid and assigned a value of 0.5. If the minimum distance dmin is greater than or equal to the preset minimum distance threshold, it is marked as a high-security grid and assigned a value of 0. This results in a grid map of impassable and passable grids.

[0095] The above analysis lays the foundation for subsequent path planning, ensuring that the searched path avoids both dynamic and static obstacles, conforms to the equipment's movement capabilities, and has a clear distinction in safety levels.

[0096] By quantifying the "no-entry zones" of dynamic obstacles (combining predicted trajectories and safe buffer distances), all "absolutely impassable" spaces can be accurately eliminated. Then, the boundaries of the zones are defined based on the current location of the device and the navigation target point, which is equivalent to building a "limited and effective" framework for subsequent path search.

[0097] Example 3: Constructing a cost function based on a raster map with both impassable and passable paths: ;

[0098] Where: α and β are the preset weighting coefficients of JL(o) ​​and QZ(o), respectively;

[0099] JL(o) ​​represents the Euclidean distance from the starting point q to the current node o;

[0100] QZ(o) represents the straight-line distance from the current node o to the navigation target point m;

[0101] DF(q) represents the dynamic risk cost of the current node q. The dynamic risk cost is equal to the risk value of the grid corresponding to node q when the device arrives at the node in the future (the arrival time is calculated based on the device's speed and distance from the node, and the time step of the obstacle's future trajectory is matched. That is, if the grid is impassable when the device arrives at the node in the future, the risk value is 1; if the grid is a passable low-safety grid when the device arrives at the node in the future, the risk value is 0.5; if the grid is a passable high-safety grid when the device arrives at the node in the future, the risk value is 0).

[0102] Further analysis based on the cost function reveals the following path search process:

[0103] SS1: Initialize the open list (stores nodes to be searched) and the closed list (stores nodes that have been searched), add the starting point m to the open list, and calculate the cost function value of the starting point m;

[0104] SS2: Select the node with the smallest cost function value from the open list as the current node o, and move it from the open list to the closed list;

[0105] SS3: Generate neighboring nodes o+1 of the current node o (within the forward direction and the device's allowed turning angle range), and analyze each neighboring node o+1:

[0106] If the adjacent node o+1 is impassable or already in the closed list, skip that node;

[0107] Calculate the cost function value of the neighboring node o+1. If the neighboring node is not in the open list, add it to the open list and record "current node" as its parent node.

[0108] If the neighboring node o+1 is already in the open list, compare the newly calculated cost function value of the neighboring node o+1 with the original cost function value. If the new cost function value is less than the original cost function value, then update the parameter information of the neighboring node o+1, including the cost function value, JL, etc.

[0109] Repeat steps SS2-SS3 until the endpoint is added to the closed list. Backtrack the parent node to obtain the obstacle avoidance planning path. If the open list is empty and the endpoint is not added to the closed list, it means there is no feasible path. Output a planning failure signal and respond to the planning failure signal to trigger the equipment's emergency braking.

[0110] In summary, multi-sensor omnidirectional perception of dynamic obstacles is achieved, providing a rich and reliable data foundation for subsequent processing. Simultaneously, data fusion and model prediction are used to predict the motion of dynamic obstacles, fully considering their motion inertia (historical state sequence), making their trajectory prediction closer to real physical laws. Furthermore, by "delineating the final obstacle avoidance target area," the global path search problem is transformed into a local but sufficiently large spatial search problem, significantly reducing the number of grids to be processed and the computational load, meeting the stringent real-time requirements of mobile devices. Moreover, by constructing a cost function, the system avoids areas that obstacles may occupy in the future, thereby generating obstacle avoidance planning paths, which helps improve path safety.

[0111] The threshold is set for comparative analysis of results to determine whether they are good or bad. The value of the threshold is determined by a combination of large-scale model analysis of sample data and human experience. It can also be adjusted appropriately based on seasonal or common-sense influencing factors.

[0112] The size of the coefficient is a specific value obtained by quantifying each parameter to facilitate subsequent comparison. The size of the coefficient depends on the amount of sample data and the corresponding operating coefficient initially set by those skilled in the art for each set of sample data; as long as it does not affect the proportional relationship between the parameter and the quantified value.

[0113] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A dynamic obstacle prediction and obstacle avoidance path planning method based on multi-sensor fusion, characterized in that, Includes the following steps: Step 1: Acquire dynamic obstacle information, location information, and device motion status information in the environment through multiple sensors; Step 2: Preprocess the multi-source sensor data collected in Step 1 to obtain standardized data; Step 3: Use a preset fusion model to fuse the standardized data obtained in Step 2 and output the fusion perception result; Step 4: Dynamic obstacle motion prediction: Based on the fusion perception results output in Step 3, retrieve the preset prediction model, input the historical motion state sequence of the obstacle, and obtain the motion trajectory information of the dynamic obstacle within the prediction time step T. Step 5: Obtain the prediction time step T and the discrete trajectory dataset; Step Six: Construction and analysis of static and dynamic restricted areas and acquisition of raster maps showing impassable and passable zones; Step 7: Plan the obstacle avoidance path using a cost function to obtain the obstacle avoidance plan path; The construction and analysis process of the static + dynamic restricted area: Obtain the basic attribute data of the dynamic obstacle. The basic attribute data includes obstacle type and size parameters. Size parameters: length L, width W, height H; Constructing scene bounding boxes for different devices based on discrete trajectory datasets: Dynamic obstacles construct a two-dimensional bounding box for ground equipment, centered at (xt, yt), with boundaries x∈[xt-L / 2, xt+L / 2] and y∈[yt-W / 2, yt+W / 2]. Dynamic obstacles construct a three-dimensional bounding box for aerial equipment, with an additional z-axis boundary, z∈[zt-H / 2, zt+H / 2]; Obtain the final safe buffer distance Da = max (maximum device outline size × preset multiple, maximum travel distance with preset response time). Extend the scene bounding box of each discrete time point t outward along all boundaries by a final safe buffer distance Da to form a dynamic restricted area at that discrete time point t, and add a timestamp label t to the dynamic restricted area; By acquiring the position and outline information of fixed obstacles through multiple sensors, a static restricted area is constructed, and then spatially superimposed with the dynamic restricted area to obtain all static + dynamic restricted areas; The process of obtaining the impassable and passable raster maps: With the current location q (xq, yq, zq) of the current device as the center, construct a circular / spherical area with a preset radius of R1 as the starting passable range, and with the navigation target point m (xm, ym, zm) as the center, construct a circular / spherical area with a preset radius of R2 as the ending passable range; Calculate the external tangent line / external plane of the passable range at the starting point and the passable range at the ending point. Based on the external tangent line / external plane, expand the width Wk = preset R1 + final safety buffer distance Da to both sides to form a closed polygon / polyhedron region, which is the initial boundary of the obstacle avoidance target area.

2. The method of claim 1, wherein the method further comprises: The process of obtaining the prediction time step T and the discrete trajectory dataset: Acquire the current motion parameters of dynamic obstacles, and simultaneously read the current mobile device's driving speed v2 and the effective detection distance S of the multi-sensor; The predicted time step T is calculated using the formula: (effective detection distance S + preset initial safety buffer distance) / min(v1, v2). If there is a decimal, it is rounded up to the nearest integer, and T is limited to [1s, 5s]. Further analysis of driving speed v2 and obstacle speed v1 shows that if driving speed v2 is greater than or equal to the preset driving speed threshold or obstacle speed v1 is greater than or equal to the preset obstacle speed threshold, then T∈[1s,6s]. In the prediction time step T, n discrete time points t are uniformly selected, t = 1, 2, ..., n, where n = T / Δt, and Δt is the preset time interval; Output the predicted position coordinates (xt, yt, zt) and motion state change of the dynamic obstacle at each discrete time point t. If the dynamic obstacle is a ground-based dynamic obstacle, then zt = 0, forming a discrete trajectory dataset with time and position correlation.

3. The method of claim 1, wherein the method further comprises: Check if the initial boundary includes all prohibited areas. If so, generate a normal signal; otherwise, generate an abnormal signal. When a normal signal is generated, the final target area boundary is determined, and the final obstacle avoidance target area is obtained. The final obstacle avoidance target area is divided into P×K or P×K×V grids according to the preset grid size, and a unique index is assigned to each grid. Initial marking: Traverse all grids. If the center of a grid falls within a restricted area, mark it as impassable and assign the corresponding grid a value of 1. Secondary labeling: For unlabeled grids, calculate the minimum distance dmin from their center to all restricted areas. Based on the minimum distance dmin, determine whether a grid is a low-security grid that allows passage and assign it a value of 0.5, or a high-security grid that allows passage and assign it a value of 0. This will result in a grid map of grids that are both impassable and passable.

4. The dynamic obstacle prediction and obstacle avoidance path planning method based on multi-sensor fusion according to claim 1, characterized in that, The obstacle avoidance planning path acquisition and analysis process: Construct a cost function based on a raster map that is both impassable and passable: ; Where: α and β are the preset weight coefficients of JL(o) ​​and QZ(o), respectively; JL(o) ​​represents the Euclidean distance from the starting point q to the current node o; QZ(o) represents the straight-line distance from the current node o to the navigation target point m; DF(o) represents the dynamic risk cost of the current node o.

5. The dynamic obstacle prediction and obstacle avoidance path planning method based on multi-sensor fusion according to claim 4, characterized in that, Further analysis based on the cost function: SS1: Initialize the open and closed lists, add the starting point m to the open list, and calculate the cost function value of the starting point m; SS2: Select the node with the smallest cost function value from the open list as the current node o, and move it from the open list to the closed list; SS3: Generate the neighboring nodes o+1 of the current node o, and analyze each neighboring node o+1: If the adjacent node o+1 is impassable or already in the closed list, skip that node; Calculate the cost function value of the neighboring node o+1. If the neighboring node is not in the open list, add it to the open list and record "current node" as its parent node. If the neighboring node o+1 is already in the open list, compare the newly calculated cost function value of the neighboring node o+1 with the original cost function value. If the new cost function value is less than the original cost function value, then update the parameter information of the neighboring node o+1. Repeat steps SS2-SS3 until the endpoint is added to the closed list. Backtrack the parent node to obtain the obstacle avoidance planning path.

Images