IoT-based high-precision obstacle avoidance and navigation system for intelligent drones

Abstract

This invention discloses a high-precision obstacle avoidance and navigation system for intelligent unmanned aerial vehicles (UAVs) based on the Internet of Things (IoT), relating to the field of intelligent UAV technology. The system includes a demand segmentation module, an intelligent obstacle avoidance module, and a display and navigation module. Its key technical features are: acquiring UAV landing requirements and abstracting and modeling these requirements as a composite topological space, with each layer containing multiple waiting points; configuring an obstacle avoidance strategy for each waiting point; identifying the minimum distance between multiple waiting points in the composite topological space and physical obstacles; if the minimum distance triggers an alarm threshold, introducing an improved particle swarm optimization algorithm to perform multi-level multi-target analysis, changing the obstacle avoidance strategy in the composite topological space, and imposing constraints by retrieving real-time wind field data; retrieving target landing sites, performing collision verification in conjunction with the updated obstacle avoidance strategy, and displaying the final landing site distribution; this invention reduces the interruption rate of flight missions in complex operational scenarios.

Description

Technical Field

[0001] This invention relates to the field of intelligent unmanned aerial vehicle (UAV) technology, specifically to a high-precision obstacle avoidance and navigation system for intelligent UAVs based on the Internet of Things (IoT). Background Technology

[0002] The application of intelligent drones in logistics delivery, power line inspection and emergency rescue is becoming increasingly widespread. The popularization of IoT collaborative sensing technology has provided a technical foundation for drone swarms to achieve distributed environmental perception and global collaborative obstacle avoidance, and has also put forward higher requirements for the architecture design, real-time performance of algorithms and dynamic environment adaptability of obstacle avoidance navigation systems.

[0003] However, in complex scenarios such as high-rise building clusters, dense landing sites, or complex terrain in the wild, existing intelligent UAV obstacle avoidance and navigation systems still face the following technical bottlenecks: On the one hand, most existing technologies focus on single-point obstacle avoidance during the flight of intelligent UAVs. For waiting point allocation and corresponding hierarchical obstacle avoidance strategies in multi-aircraft swarm operation scenarios, single-UAV planning modes are often adopted. In the process of generating feasible navigation paths, methods such as artificial potential field methods and fast random trees are often used. When faced with dynamic physical obstacles and multiple constraints, these methods are prone to getting stuck in local optima and lack real-time performance; especially when considering the conditions of multi-aircraft paths. In low-altitude scenarios, multi-level path conflicts are prone to occur, and the safety redundancy of waiting points is insufficient, making it unable to adapt to the requirements of collaborative obstacle avoidance. On the other hand, gusts of wind and sudden moving obstacles in low-altitude scenarios can easily cause pre-planned obstacle avoidance strategies to fail. Although there are obstacle avoidance planning methods based on particle swarm optimization, most of them are fixed target optimization modes. In high-risk scenarios, if the same target optimization mode is used, it will be difficult to achieve fast convergence to a certain extent, reducing the robustness of the model. Especially in the landing phase, UAV position drift, untimely obstacle avoidance, and insufficient landing accuracy are likely to occur, increasing the execution interruption rate of flight missions. Summary of the Invention

[0004] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a high-precision obstacle avoidance and navigation system for intelligent unmanned aerial vehicles (UAVs) based on the Internet of Things (IoT). Through a demand segmentation module, an intelligent obstacle avoidance module, and a display and navigation module, the UAV landing requirements are abstractly modeled as a composite topological space. An obstacle avoidance strategy is configured for each waiting point in the composite topological space. By setting a rule engine, the system sequentially performs feasibility assessments on the current flight mission to obtain feasibility results. By monitoring performance stability events, it identifies situations where multiple waiting points and physical obstacles trigger multi-level alarm thresholds. Through multi-level, multi-objective analysis, the obstacle avoidance strategy is modified, and real-time wind field data is introduced as a mandatory constraint. The generated optimization results are then further filtered and trained, thus solving the problems mentioned in the background technology.

[0005] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: This application provides a high-precision obstacle avoidance navigation system for intelligent unmanned aerial vehicles based on the Internet of Things, the system comprising: The requirement segmentation module obtains the drone landing requirements and abstracts and models them into a composite topology space. Each layer of the composite topology space contains multiple waiting points. An obstacle avoidance strategy is configured for each waiting point, including navigation path, track deviation angle, track tilt angle, cruise speed, and minimum turning radius. The composite topology space includes a gathering layer, an upper waiting layer, and a lower waiting layer. The intelligent obstacle avoidance module identifies the minimum distance between multiple waiting points in the composite topology space and physical obstacles. If the minimum distance triggers an alarm threshold, an improved particle swarm algorithm is introduced to perform multi-level multi-target analysis, change the obstacle avoidance strategy in the composite topology space, and enforce constraints by retrieving real-time wind field data. The display navigation module retrieves the target landing site, performs collision verification in conjunction with the updated obstacle avoidance strategy, and displays the final landing site distribution.

[0006] Furthermore, obtain drone landing requirements, including: Respond to multi-level operation commands, wherein the multi-level operation commands include at least an operation command ID, an operation command type, and an operation command content, and the command type includes a release command, a take-off command, and a landing command, and the operation command content includes at least the UAV landing requirements; The system parses multi-level operation commands to generate multiple flight mission packages, and each flight mission package includes at least: payload, remaining battery power, target landing site, and real-time UAV position; the target landing site includes a primary landing site and multiple secondary landing sites. Multi-level operation instructions are imported into a preset rule engine. The feasibility of the current flight mission is evaluated sequentially through the first rule and the second rule to obtain the feasibility results, including the first level result and the second level result. The first rule includes the judgment of the airspace boundary and the performance level matching of each UAV, and the performance level includes three levels: high, medium and low. The second rule includes the judgment of inter-aircraft collision avoidance. Based on the feasibility results, the initial distribution of topological nodes in the composite topological space is generated.

[0007] Furthermore, feasibility results are obtained, including: Determine whether the multi-level execution instructions meet the first rule to obtain the first evaluation result; If the first evaluation result meets the first rule, determine whether the multi-level operation instructions meet the second rule and obtain the second evaluation result; if the first evaluation result does not meet the first rule, output the second degree result and determine that the flight mission is not feasible. The first rule is based on the following criteria: Airspace Boundaries: Introduce a four-level geofence, retrieve the obstacle avoidance strategy of the current task level, analyze the topological distance between the navigation path of each waiting point and the geofence at each level, and determine the access permissions of the next task level. Performance level matching: retrieve flight mission packages to construct a three-dimensional dynamic envelope and perform three-dimensional spatial matching judgment; integrate the identified performance stability events with the three-dimensional dynamic envelope into a target dataset, extract the mean and standard deviation of performance margins, establish the probability estimation envelope boundary of the current mission level, and define it as a standard envelope. If the second evaluation result meets the second rule, output the first-level result, determining that the task is feasible and allowing execution; if the second evaluation result does not meet the second rule, output the second-level result. The second rule is based on the following criteria: Collision avoidance judgment between aircraft: retrieve the obstacle avoidance strategy at the current mission level, analyze the navigation track deflection angle, track inclination angle, cruise speed and minimum turning radius of each waiting point; predict the track line of each waiting point in the next N time moments, identify the conflict point through the intersection algorithm, reconstruct the navigation path according to the minimum turning radius, and force the corresponding waiting point to deviate from the current track line; where N is a positive integer.

[0008] Furthermore, the step of generating the target landing site also includes: Based on the primary landing site, an obstacle avoidance deviation region is constructed. The obstacle avoidance deviation region is divided into equally spaced grids to form several grid cells, which serve as candidate secondary landing sites. Retrieve drone sensor parameters, analyze the occupancy score of each grid cell, determine the occupancy status of the grid cell based on the occupancy score, including occupied and unoccupied status; generate a real-time occupancy probability map based on the occupancy status, and identify physical obstacles in obstacle avoidance deviation areas by threshold comparison; Boolean logic operations are introduced to filter unoccupied grid cells from the grid cells to obtain the effective landing area; candidate secondary landing sites located in the effective landing area are then selected and marked as the final secondary landing sites. The drone sensing parameters include real-time depth maps, 3D point clouds, and static ground masks.

[0009] Furthermore, an improved particle swarm optimization algorithm is introduced to perform multi-level, multi-objective analysis, including: Alarm thresholds include primary alarm thresholds and secondary alarm thresholds; Triggering the Level 2 alarm threshold, perform single-node multi-objective optimization: collect all waiting points that trigger the Level 2 alarm threshold, extract the corresponding real-time UAV position for each waiting point and mark it as the first position vector, randomly initialize the particle swarm in the first search space of the waiting point; with the optimization objectives of minimizing the navigation path deviation and maximizing the obstacle avoidance safety distance of the waiting point, perform optimization update; Triggering the Level 1 alarm threshold triggers global multi-objective optimization: Collect the set of waiting points in the assembly layer, upper waiting layer, and lower waiting layer under the triggering of the Level 1 alarm threshold, extract the corresponding real-time UAV positions, and mark them as the second position vector. Randomly initialize the particle swarm in the second search space. With the optimization objectives of minimizing the spatial topology distribution uniformity of all waiting points, minimizing the rate of change of total cruise speed, and maximizing the obstacle avoidance safety distance, perform optimization update.

[0010] Furthermore, the updated obstacle avoidance strategy is subject to mandatory constraints by retrieving real-time wind field data, including: The association between topological nodes and links in the composite topological space is marked; where the topological node is the waiting point in the composite topological space, the link is the navigation path generated by the improved particle swarm optimization algorithm, and each link carries real-time wind field data; where the real-time wind field data includes wind speed and wind direction; Set the functional relationship between the navigation path and real-time wind field data. By obtaining the solutions of each waiting point on the navigation path generated by optimization, a wind field feature matrix is ​​formed. Perform Boolean operation on the wind field feature matrix and the standard feature matrix, filter out the wind field feature matrices that cannot meet the first rule, mark them as 0, trigger the forced constraint, immediately discard the current optimization result, and re-optimize and update.

[0011] Furthermore, the obstacle avoidance strategy is modified, including: modifying the command type, modifying the content set, and modifying the relationship set; wherein, the command type is the type of execution command; the content set is the UAV landing requirements; and the relationship set is the relationship between topology nodes.

[0012] Further, collision checking is performed, including: Retrieve the target landing site and correlate it with the updated navigation path in time and space to establish a dynamic safety envelope; Based on the target landing site, the effective landing area is extracted. Collision verification is performed on the dynamic safety envelope within the effective landing area. Cases where the dynamic safety envelope intersects with physical obstacles are filtered out, and the corresponding landing sites are removed. After the target landing site completes the collision verification, the final landing site distribution is output and displayed.

[0013] (III) Beneficial Effects This invention provides a high-precision obstacle avoidance navigation system for intelligent unmanned aerial vehicles based on the Internet of Things, which has the following beneficial effects: This invention abstracts and models the landing requirements of drones into a composite topological space containing a gathering layer, an upper waiting layer, and a lower waiting layer, thus achieving hierarchical management of intelligent drone swarms. Combined with preset obstacle avoidance strategies for waiting points, it can effectively handle high-density landing tasks and adapt to drone operation scenarios of different scales. By importing flight mission packages into a preset rule engine, after passing through the first and second rules, it first performs a judgment on the matching between the composite airspace boundary and the performance level of each drone. Under the corresponding conditions, it performs a judgment on inter-drone collision avoidance. By monitoring performance stability events and utilizing this hierarchical rule engine, in the subsequent analysis of particle swarm optimization, it can not only be independently updated and iterated, improving reusability, but also significantly reduces the probability of crashes caused by performance insufficiency events, greatly improving safety. This invention sets tiered alarm thresholds based on the minimum distance between waiting points and obstacles, and triggers differentiated improved particle swarm optimization strategies for different alarm thresholds. The first-level alarm threshold performs global multi-objective optimization to ensure path optimality, while the second-level alarm threshold performs single-node fast optimization to ensure response speed. After iterative convergence, the particles with the lowest overall fitness are selected as the globally optimal waiting point set, and the verification of the first and second rules is re-executed. At the same time, real-time wind field data is retrieved to further enforce constraints, and the optimization results generated by the improved particle swarm algorithm are further filtered or trained to ensure the accuracy and orderliness of the final result. This invention retrieves the target landing site and performs collision verification in conjunction with an updated obstacle avoidance strategy, eliminating landing sites where the dynamic safety envelope intersects with physical obstacles. This approach ensures the optimality and safety of the primary landing site to a certain extent. At the same time, with the addition of visualization output, it is adapted to fully automatic landing execution by intelligent UAVs, greatly reducing the interruption rate of flight missions. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the modules of the present invention. Detailed Implementation

[0015] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0016] The core of this invention lies in: abstracting and modeling the landing requirements of UAVs into a composite topology space through a demand segmentation module, an intelligent obstacle avoidance module, and a display and navigation module; configuring obstacle avoidance strategies for each waiting point in the composite topology space; performing feasibility assessments on the current flight mission sequentially by setting a rule engine to obtain feasibility results; identifying situations where multiple waiting points and physical obstacles trigger multi-level alarm thresholds by monitoring performance stability events; changing obstacle avoidance strategies through multi-level and multi-objective analysis; introducing real-time wind field data for mandatory constraints; and further filtering and training on the generated optimization results to reduce the interruption rate of flight missions.

[0017] Example: This invention provides a high-precision obstacle avoidance and navigation system for intelligent unmanned aerial vehicles (UAVs) based on the Internet of Things (IoT). Figure 1 This is a schematic diagram of the modules of the present invention; please refer to it. Figure 1 The system includes: a demand segmentation module, an intelligent obstacle avoidance module, and a display and navigation module, and the demand segmentation module, the intelligent obstacle avoidance module, and the display and navigation module are connected to each other. The following is an explanation of each module: Demand segmentation module: Obtain UAV landing requirements and abstract them into a composite topology space. Each layer of the composite topology space contains multiple waiting points. Configure obstacle avoidance strategies for each waiting point, including navigation path, track deviation angle, track tilt angle, cruise speed, and minimum turning radius. The composite topology space includes a gathering layer, an upper waiting layer, and a lower waiting layer. Obtaining drone landing requirements includes: receiving and responding to multi-level operational instructions from the control center via an IoT interface; wherein, the multi-level operational instructions include operational instruction ID, operational instruction type, operational instruction content, and response timestamp, and the operational instruction type includes release instructions, takeoff instructions, and landing instructions, and the operational instruction content includes drone landing requirements, which include, but are not limited to, the target landing point and its coordinates and the expected landing scale; parsing the multi-level operational instructions based on the TF-IDF algorithm, and by extracting keywords of the operational instruction type, i.e., identifying keywords for release, takeoff, and landing, constructing three task levels, including release level, takeoff level, and landing level, generating corresponding flight mission packages for different taskers, and each flight mission package including, but not limited to, payload, remaining battery power, target landing site, and real-time drone position; wherein, the target landing site includes a primary landing site and multiple secondary landing sites; Simultaneously, taking the target landing point's coordinates as the center, this analysis focuses on the main landing site, dividing it into three layers of topology space in the vertical dimension. Based on the three mission levels, a composite topology space is constructed, consisting of an assembly layer, an upper waiting layer, and a lower waiting layer. The assembly layer receives incoming UAVs from various routes and assigns them to preset waiting points for initial sorting and flow adjustment. It has a large horizontal radius to provide ample buffer space and represents the transition airspace. The upper waiting layer further buffers and adjusts the incoming flow and is located below the assembly layer. The lower waiting layer performs final incoming sorting and prepares for subsequent descent and is located below the upper waiting layer. The upper and lower waiting layers are two ring-shaped waiting layers, not shown in the figure. Cross-layer connections are established to link the release levels. The system consists of a takeoff-level topology node and a landing-level topology node, connecting the takeoff-level and landing-level topology nodes, with each level operating independently. Within each layer of the composite topology space, multiple waiting points are generated through spatial discretization sampling based on the expected landing scale. Each waiting point is assigned a unique spatial index and an occupancy flag. The occupancy flag includes: 0 for free (unoccupied, allowing the drone to fly through), and 1 for occupied (occupied, preventing the drone from flying through and requiring obstacle avoidance). Drones awaiting landing are mapped to waiting points sequentially, and the dwell time of each drone is monitored in real-time. If the dwell time exceeds a safe threshold, the drone's task level is forcibly updated, granting access to the next task level. The step of generating the target landing site also includes: constructing an obstacle avoidance deviation region based on the primary landing site, dividing the obstacle avoidance deviation region into equally spaced grids to form several grid cells as candidate secondary landing sites; retrieving UAV sensor parameters, including: real-time depth maps obtained through image sensors, 3D point clouds obtained through LiDAR, and static ground masks synchronized by IoT edge nodes; it should be noted that the above sensors or devices are not shown in the figure, but are adaptively installed to ensure the collection of corresponding UAV sensor parameters and to synchronize the collected UAV sensor parameters in time to ensure subsequent processing and analysis; A semantic segmentation algorithm is used to extract features from the real-time depth map. By classifying pixels one by one, the proportion of pixels representing physical obstacles is obtained. The proportion of pixels representing physical obstacles is defined as the ratio of the number of pixels labeled as physical obstacles to the total number of pixels within a preset pixel window. Based on the proportion of pixels representing physical obstacles, the Canny edge detection algorithm is used to extract edge contour features. Combined with the category mask output by the semantic segmentation algorithm, noisy edges are removed to generate a set of boundary coordinates corresponding to continuous closed physical obstacles. At the same time, a boundary region is constructed based on the boundary coordinate set, and the depth map within the boundary region is collected for bilinear interpolation to identify the boundaries of physical obstacles in the obstacle avoidance deviation area. The semantic segmentation algorithm can be a pre-trained lightweight BiSeNetV2 semantic segmentation network. In this embodiment, the semantic segmentation network is based on TensorRT and performs INT8 quantization. Its training dataset contains depth maps labeled under different UAV operation scenarios. The labeled categories include, but are not limited to, physical obstacles, which can specifically be represented as thin power lines, buildings, trees, glass curtain walls, and reflective surfaces. By combining 3D point cloud spatial mapping with coordinate space transformation through the camera intrinsic parameter matrix, pixel coordinates are mapped to the coordinate system of the 3D point cloud, achieving spatial mapping between the depth map and the 3D point cloud, and matching them to the corresponding grid cells. Simultaneously, physical obstacles from the static ground mask are mapped to the corresponding grid cells. An occupancy score for each grid cell is calculated according to a preset occupancy determination rule. This rule is based on extracting the height standard deviation and the percentage of physical obstacle pixels from the 3D point cloud of the grid cell, and generating an occupancy score through weighted summation. This involves first normalizing the height standard deviation and the percentage of physical obstacle pixels to eliminate dimensional influences, ensuring the subsequent formula construction has physical meaning. Then, the normalized height standard deviation and the first weighted summation are used to calculate the occupancy score. The process involves a multiplication process: the normalized percentage of physical obstacle pixels is multiplied by the second weight, and the results are summed to generate an occupancy ratio: normalized percentage of physical obstacle pixels × first weight + normalized percentage of physical obstacle pixels × second weight. Finally, the occupancy score is compared to a standard score threshold: if the occupancy score is less than or equal to the standard score threshold, the area is considered unoccupied; otherwise, if the occupancy score is greater than the standard score threshold, the area is considered occupied. A real-time occupancy probability map is generated based on the occupancy status. Boolean logic is introduced to perform an XOR operation between all grid cells and the real-time occupancy probability map, removing occupied grid cells and constructing an effective landing area. Candidate secondary landing sites located within the effective landing area are selected and marked as the final secondary landing sites. Both the first and second weights are dynamic values, determined using a genetic algorithm: a high-dimensional target space is set, and an initial population is randomly generated, consisting of multiple particles; during initialization, a certain number of individuals are randomly generated and labeled as a weight combination of the first and second weights, ensuring that all weights are within a suitable range, set between 0 and 1, and that the sum of the weights is 1; based on individual fitness, individuals with high fitness are selected as parents, and a roulette wheel selection strategy is used to sequentially select particles from the population and place them into the mating pool until the number of particles in the mating pool reaches the maximum. New individuals are obtained and added to the population through selection, crossover, and mutation operations to update the population composition. It should be noted that during crossover: a crossover point is selected, the parent's genes are split into two parts at that point, and then the two parts are exchanged to generate two offspring. Furthermore, the crossover operation is not performed on every parent pair; a crossover probability is usually set, for example, crossover occurs with a 60% probability, while the parents remain unchanged with a 40% probability. After updating the population, if the average fitness or the optimal solution no longer changes significantly, the result is output, and training stops. The standard score threshold is determined based on offline simulation sampling. The Monte Carlo algorithm is used to simulate the data collection of sensor parameters from an intelligent drone under different lighting and altitude conditions, including real-time depth maps, 3D point clouds, and static ground masks, generating a large number of sample data pairs. The occupancy status of each grid cell is labeled, with the occupancy score of each grid cell as the test variable and the occupancy status of each grid cell as the state variable. Candidate thresholds in the 0-1 interval are traversed, and the true positive rate and false positive rate corresponding to each candidate threshold are statistically analyzed. An ROC curve is plotted with the false positive rate on the x-axis and the true positive rate on the y-axis. Simultaneously, extract the occupancy score corresponding to the maximum value of the Youden index and mark it as the initial optimal critical threshold; traverse all sample data pairs, and statistically analyze the four statistical values ​​of true positive, false negative, true negative, and false positive under this threshold to generate the corresponding confusion matrix; in the occupied grid cells, calculate the proportion correctly identified as occupied based on the confusion matrix and mark it as the true positive rate; in the unoccupied grid cells, calculate the proportion correctly identified as unoccupied and mark it as the true negative rate. That is, each data point on the ROC curve corresponds to an initial true positive rate and true negative rate.

[0018] By abstracting and modeling the landing requirements of drones into a composite topological space containing a gathering layer, an upper waiting layer, and a lower waiting layer, hierarchical management of intelligent drone swarms is achieved. Combined with the preset obstacle avoidance strategy of waiting points, it can effectively cope with high-density landing tasks and adapt to drone operation scenarios of different scales.

[0019] The flight mission package is imported into the preset rule engine. The feasibility of the current flight mission is evaluated sequentially by the first rule and the second rule to obtain the feasibility result. The feasibility result includes the first result and the second result. The specific steps of performing the feasibility evaluation of the current flight mission sequentially to obtain the feasibility result include: determining whether the multi-level operation instructions meet the first rule to obtain the first evaluation result. If the first evaluation result meets the first rule, determine whether the multi-level operation instructions meet the second rule and obtain the second evaluation result; if the first evaluation result does not meet the first rule, output the second degree result and determine that the flight mission is not feasible. The first rule includes: determining the matching between airspace boundaries and the performance levels of each UAV; among which, the performance levels include three levels: high, medium, and low. Airspace Boundaries: A four-level geofencing system is introduced, including four levels: restricted area, flight boundary, buffer zone, and permitted area. The restricted area is an area such as the core area of ​​an airport runway; the flight boundary is a regular flight boundary or height-restricted area; the buffer zone is a buffer zone or flow control zone; and the permitted area is used for switching between mission levels. The system retrieves the obstacle avoidance strategy of the current mission level and analyzes the topological distance between the navigation path of each waiting point and the geofence at each level. This topological distance is set as the minimum Euclidean distance between the waiting point on the navigation path and the boundary of each geofence. Based on the calculated topological distance, the system determines whether the UAV has permission to enter the next stage. For example, if the topological distance between the navigation path and the geofence corresponding to the permitted area is less than a preset topological threshold and maintains a safe margin with the restricted area, then the permission is satisfied, and the system activates the permission switch for the next mission level, such as allowing a switch from the upper waiting level to the lower waiting level. If the topological distance of the navigation path intrudes into the flight boundary due to avoiding physical obstacles, then the permission is not satisfied, and the intelligent UAV is required to continue circling at the current mission level until the airspace is safe again. Performance level matching: The flight mission package is retrieved to construct a three-dimensional dynamic envelope. A three-dimensional spatial matching judgment is performed. Specific steps include: extracting the payload and remaining battery power from the flight mission package; mapping the payload as the horizontal axis, the remaining battery power as the vertical axis, and time as the third axis to a three-dimensional dynamic envelope at the same moment; superimposing the three-dimensional dynamic envelope with the standard envelope to obtain the total overlapping area and the total non-overlapping area on the plane formed by the horizontal and vertical axes; and generating a performance margin based on the combination of the overlapping and non-overlapping areas. The calculation formula is: Performance Margin = [Total Overlapping Area / (Total Overlapping Area + Total Non-Overlapping Area)] The formula is calculated as follows: [[product] × margin factor]; where the margin factor is a predefined value, ranging from 0 to 1. During the calculation process, the overlapping and non-overlapping total areas need to be normalized to eliminate the influence of dimensions and ensure the meaning of the formula calculation. The larger the performance margin value, the more stable the performance of the intelligent drone, corresponding to the load and remaining power being within the safety margin, the higher the system robustness, and it is judged as a performance stability event. The smaller the performance margin value, the more unstable the performance of the intelligent drone, corresponding to the load and remaining power possibly exceeding the safety margin, which is easy to trigger a performance insufficiency event. The performance margin is compared with the performance margin range: If the performance margin is less than the performance margin range, the current performance margin is marked as 'a', assigned a first-level character, and 'a' is combined with the first-level character to generate a low performance level; simultaneously, the low performance level is marked as a performance insufficiency event. If the performance margin is within the performance margin range and includes the range boundary, the current performance margin is marked as 'b', assigned a second-level character, and 'b' is combined with the second-level character to generate a medium performance level. If the performance margin is greater than the performance margin range, the current performance margin is marked as 'c', assigned a third-level character, and 'c' is combined with the third-level character to generate a medium performance level. The system identifies high-performance levels and marks medium and high-performance levels as performance-stable events. It integrates the identified performance-stable events with the three-dimensional dynamic envelope into a target dataset, extracts the mean and standard deviation of the performance margin, and defines the average performance margin as the mean of the performance margin plus a multiple of the standard deviation. It then constructs a probability envelope boundary based on confidence intervals within the current task level, maps this boundary to three-dimensional space, and generates a physical standard envelope as a performance comparison benchmark through boundary closure processing. It should be noted that the value of the multiple of 2 is just an example; the specific setting should be based on the actual situation and will not be elaborated upon here. If the second assessment result meets the second rule, the first-level result is output, and the mission is deemed feasible and approved for execution; if the second assessment result does not meet the second rule, the second-level result is output, and the flight mission is deemed infeasible. The second rule includes: Collision avoidance judgment between aircraft: Retrieve the obstacle avoidance strategy at the current task level, analyze the navigation track deflection angle, track inclination angle, cruising speed, and minimum turning radius of each waiting point; analyze the three-dimensional curve formed by the navigation track deflection angle, track inclination angle, and cruising speed to predict the track line of each waiting point within the next N time moments; where N is a positive integer; send the predicted track line to the rule engine in real time, perform intersection calculation with the predicted track lines of neighboring waiting points, and search for whether there is geometric overlap between the track lines; for example: identify at least two waiting points whose predicted track lines exist in space, with a spatial distance less than a preset safety distance threshold, and whose corresponding arrival time difference is within the coordinate set of a preset collision window, identifying potential conflict points; execute an avoidance heading based on the minimum turning radius, identify the currently existing waiting points, and for any waiting point, obtain the waiting... The system calculates the real-time position of the drone and performs spatial offset mapping. By drawing a circle with the minimum turning radius, an arc is formed, representing the fastest maneuver trajectory that the intelligent drone can execute. By generating an avoidance tangent line along the circumference of the arc at its end, it ensures that the new waiting point generated at its end does not spatially overlap with other waiting points in the same airspace (or the predicted trajectory of other drones). The drone is then forcibly guided to the end of the avoidance tangent line, completing the navigation path reconstruction of the waiting point and achieving avoidance. In addition, if the trajectory of drones P and Q is detected to intersect, the rule engine first analyzes the minimum turning radius of drone P. If drone P cannot achieve rapid avoidance by changing the navigation trajectory angle at the current cruising speed, the system will issue a setting command to drone Q, increasing the trajectory tilt angle of drone Q to make it climb rapidly, thereby avoiding the conflict in the vertical dimension.

[0020] By importing flight mission packages into a pre-defined rule engine, and following the first and second rules, the engine first performs a judgment on the matching of the composite airspace boundary and the performance level of each UAV. Under the corresponding conditions, it then performs a judgment on inter-UAV collision avoidance. Through this hierarchical rule engine, in the subsequent analysis of particle swarm optimization, it can not only be updated and iterated independently, improving reusability, but also significantly reduce the probability of crashes caused by performance deficiencies, greatly improving safety.

[0021] Intelligent obstacle avoidance module: Identifies the minimum distance between multiple waiting points in the composite topology space and physical obstacles. If the minimum distance triggers the alarm threshold, an improved particle swarm algorithm is introduced to perform multi-level multi-target analysis, change the obstacle avoidance strategy in the composite topology space, and enforce constraints on the updated obstacle avoidance strategy by retrieving real-time wind field data. The system acquires multiple waiting points and their coordinates at each level within the composite topology space in real time, calculates the minimum distance between each waiting point and a physical obstacle using Euclidean distance, and presets the collision boundary radius. If the minimum distance between the waiting point and the physical obstacle is less than or equal to a multiple of the preset boundary radius, but greater than a multiple of the preset boundary radius, a secondary alarm threshold is triggered, and single-node multi-objective optimization is performed: all waiting points that trigger the secondary alarm threshold are collected. For each waiting point, the corresponding real-time UAV position is extracted and marked as the first position vector. Particle swarms are randomly initialized in the first search space of the waiting point. The first search space is represented by a spherical region with radius L1 centered on the first position vector, and the value range of L1 is set to the result of multiplying the highest cruising speed at the current performance level of the waiting point by the step size of each iteration of the particle swarm algorithm. With the optimization objectives of minimizing the navigation path deviation and maximizing the obstacle avoidance safety distance of the waiting point, optimization and updates are performed only on the trajectory deflection angle and trajectory tilt angle of the waiting point under the constraint of satisfying the minimum turning radius. The fitness function under single-node multi-objective is calculated, and the particle with the smallest comprehensive fitness is selected as the optimal waiting point for the single node. If the minimum distance between the waiting point and the physical obstacle is less than or equal to a preset boundary radius of 1.5 times, a Level 1 alarm threshold is triggered, and global multi-objective optimization is performed: The set of waiting points in the assembly layer, upper waiting layer, and lower waiting layer under the Level 1 alarm threshold is collected, the corresponding real-time UAV positions are extracted and marked as the second position vector, and the particle swarm is randomly initialized in the second search space; where the second search space is represented as the union envelope region of the assembly layer, upper waiting layer, and lower waiting layer; optimization is performed with the optimization objectives of minimizing the spatial topological distribution uniformity of all waiting points, minimizing the rate of change of total cruising speed, and maximizing the obstacle avoidance safety distance; the fitness function under global multi-objective optimization is calculated, and the particle with the smallest comprehensive fitness is selected as the globally optimal waiting point; The updated obstacle avoidance strategy is constrained by retrieving real-time wind field data, including: marking the association between topological nodes and links in the composite topology space; wherein, the topological nodes are waiting points in the composite topology space, and the links are navigation paths optimized by the improved particle swarm optimization algorithm, and each link carries real-time wind field data; wherein, the real-time wind field data includes wind speed, wind direction and air humidity; Set the functional relationship between the navigation path and real-time wind field data: within a preset time period, extract the feature vectors of wind speed and wind direction based on the real-time wind field data. At the same time, extract the cruise speed generated by the improved particle swarm optimization algorithm, and combine it with the airspeed of the intelligent drone to characterize the actual speed of the intelligent drone in the airflow; calculate the lateral drift due to wind field data interference. In the formula, Indicates the amount of lateral drift. This represents the characteristic vector composed of wind speed and wind direction. This represents the cruising speed generated by the improved particle swarm optimization algorithm. Indicates airspeed. This represents the integral operation within a preset time period from 0 to t. It should be noted that lateral drift, as a core indicator for evaluating the physical feasibility of the obstacle avoidance strategy, has the following physical meaning: predicting the probability that the intelligent UAV will deviate from the preset safety envelope due to environmental interference during obstacle avoidance maneuvers; constructing a wind field feature matrix by obtaining the solutions for each waiting point on the optimized navigation path; performing a Boolean operation between the wind field feature matrix and the standard feature matrix to filter out wind field feature matrices that cannot satisfy the first rule, and marking them as 0. 0 indicates that the current particle swarm optimization result is physically unreachable, thus driving the system to forcibly change or roll back the execution strategy; simultaneously, triggering a forced constraint, immediately discarding the current optimization result, and re-updating the optimization until the first rule is satisfied; conversely, if it is 1, it indicates that the current particle swarm optimization result is physically reachable, maintaining the original particle swarm optimization result. The obstacle avoidance strategy is modified, including: modifying the command type, modifying the content set, and modifying the relationship set. The command type is the type of execution command; the content set is the UAV landing requirements; and the relationship set is the relationship between topology nodes. It should be noted that the relationship between topology nodes can be a semantic relationship. If a certain task level needs to modify the obstacle avoidance strategy, the semantic relationships between each task level are used to accurately track the waiting points with linkages and dynamically adjust the navigation path under the condition of meeting the performance stability event.

[0022] A tiered alarm threshold is set based on the minimum distance between the waiting point and the obstacle. Differentiated improved particle swarm optimization strategies are triggered for different alarm thresholds. The first-level alarm threshold performs global multi-objective optimization to ensure path optimality, while the second-level alarm threshold performs single-node fast optimization to ensure response speed. After iterative convergence, the particle with the lowest comprehensive fitness is selected as the global optimal waiting point set, and the verification of the first and second rules is re-executed. At the same time, real-time wind field data is retrieved to further enforce constraints. The optimization results generated by the improved particle swarm algorithm are then filtered or trained again to ensure the accuracy and orderliness of the final result.

[0023] The display navigation module retrieves the target landing site, performs collision verification in conjunction with the updated obstacle avoidance strategy, and displays the final landing site distribution. The target landing site is retrieved and spatiotemporally correlated with the updated navigation path. At the same time, combined with performance level, navigation path and real-time wind field data, a corresponding three-dimensional dynamic safety envelope is constructed for each waiting point of the navigation path. At the same time, the dynamic safety envelope of each waiting point is strongly bound to the timestamp and spatial coordinates of the navigation path to ensure spatiotemporal matching. Based on the target landing site, the effective landing area is extracted. Collision verification is performed on the dynamic safety envelope within the effective landing area. If the dynamic safety envelope intersects with a physical obstacle, the corresponding landing site is removed. If the dynamic safety envelope does not intersect with a physical obstacle, the corresponding landing site is retained. After the target landing site completes the collision verification, the results are simultaneously output to the ground control platform and the UAV's onboard display terminal to complete the visualization display in the three-dimensional scene. This intuitively presents the distribution of landing sites, providing precise instructions for the fully automated landing of UAVs and clear decision support for manual emergency control.

[0024] In the application, the various formulas mentioned are all calculated by removing dimensions and taking their numerical values. The formulas are derived from the most recent real-world situation by collecting a large amount of data and conducting software simulations. The formulas are set by those skilled in the art according to the actual situation.

[0025] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0026] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0027] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A high-precision obstacle avoidance and navigation system for intelligent unmanned aerial vehicles based on the Internet of Things, characterized in that: The system includes: The requirement segmentation module obtains the drone landing requirements and abstracts and models them into a composite topology space. Each layer of the composite topology space contains multiple waiting points. An obstacle avoidance strategy is configured for each waiting point, including navigation path, track deviation angle, track tilt angle, cruise speed, and minimum turning radius. The composite topology space includes a gathering layer, an upper waiting layer, and a lower waiting layer. The intelligent obstacle avoidance module identifies the minimum distance between multiple waiting points in the composite topology space and physical obstacles. If the minimum distance triggers an alarm threshold, an improved particle swarm algorithm is introduced to perform multi-level multi-target analysis, change the obstacle avoidance strategy in the composite topology space, and enforce constraints by retrieving real-time wind field data. The display navigation module retrieves the target landing site, performs collision verification in conjunction with the updated obstacle avoidance strategy, and displays the final landing site distribution.

2. The high-precision obstacle avoidance navigation system for intelligent unmanned aerial vehicles based on the Internet of Things as described in claim 1, characterized in that, Obtain drone landing requirements, including: Respond to multi-level operation commands, wherein the multi-level operation commands include at least an operation command ID, an operation command type, and an operation command content, and the command type includes a release command, a take-off command, and a landing command, and the operation command content includes at least the UAV landing requirements; The system parses multi-level operation commands to generate multiple flight mission packages, and each flight mission package includes at least: payload, remaining battery power, target landing site, and real-time UAV position; the target landing site includes a primary landing site and multiple secondary landing sites. Multi-level operation instructions are imported into a preset rule engine. The feasibility of the current flight mission is evaluated sequentially through the first rule and the second rule to obtain the feasibility results, including the first level result and the second level result. The first rule includes the judgment of the airspace boundary and the performance level matching of each UAV, and the performance level includes three levels: high, medium and low. The second rule includes the judgment of inter-aircraft collision avoidance. Based on the feasibility results, the initial distribution of topological nodes in the composite topological space is generated.

3. The IoT-based intelligent unmanned aerial vehicle (UAV) high-precision obstacle avoidance navigation system according to claim 2, characterized in that, To obtain feasible results, including: Determine whether the multi-level execution instructions meet the first rule to obtain the first evaluation result; If the first evaluation result meets the first rule, determine whether the multi-level operation instructions meet the second rule and obtain the second evaluation result; if the first evaluation result does not meet the first rule, output the second degree result and determine that the flight mission is not feasible. The first rule is based on the following criteria: Airspace Boundaries: Introduce a four-level geofence, retrieve the obstacle avoidance strategy of the current task level, analyze the topological distance between the navigation path of each waiting point and the geofence at each level, and determine the access permissions of the next task level. Performance level matching: retrieve flight mission packages to construct a three-dimensional dynamic envelope and perform three-dimensional spatial matching judgment; integrate the identified performance stability events with the three-dimensional dynamic envelope into a target dataset, extract the mean and standard deviation of performance margins, establish the probability estimation envelope boundary of the current mission level, and define it as a standard envelope. If the second evaluation result meets the second rule, output the first-level result, determining that the task is feasible and approved for execution; if the second evaluation result does not meet the second rule, output the second-level result. The second rule is based on the following criteria: Collision avoidance judgment between aircraft: retrieve the obstacle avoidance strategy at the current mission level, analyze the navigation track deflection angle, track inclination angle, cruise speed and minimum turning radius of each waiting point; predict the track line of each waiting point in the next N time moments, identify the conflict point through the intersection algorithm, reconstruct the navigation path according to the minimum turning radius, and force the corresponding waiting point to deviate from the current track line; where N is a positive integer.

4. The high-precision obstacle avoidance navigation system for intelligent unmanned aerial vehicles based on the Internet of Things according to claim 2, characterized in that, The step of generating the target landing site also includes: Based on the primary landing site, an obstacle avoidance deviation region is constructed. The obstacle avoidance deviation region is divided into equally spaced grids to form several grid cells, which serve as candidate secondary landing sites. Retrieve drone sensor parameters, analyze the occupancy score of each grid cell, determine the occupancy status of the grid cell based on the occupancy score, including occupied and unoccupied status; generate a real-time occupancy probability map based on the occupancy status, and identify physical obstacles in obstacle avoidance deviation areas by threshold comparison; Boolean logic operations are introduced to filter unoccupied grid cells from the grid cells to obtain the effective landing area; candidate secondary landing sites located in the effective landing area are then selected and marked as the final secondary landing sites. The drone sensing parameters include real-time depth maps, 3D point clouds, and static ground masks.

5. The high-precision obstacle avoidance navigation system for intelligent unmanned aerial vehicles based on the Internet of Things according to claim 1, characterized in that, An improved particle swarm optimization algorithm is introduced to perform multi-level, multi-objective analysis, including: Alarm thresholds include primary alarm thresholds and secondary alarm thresholds; Triggering the Level 2 alarm threshold, perform single-node multi-objective optimization: collect all waiting points that trigger the Level 2 alarm threshold, extract the corresponding real-time UAV position for each waiting point and mark it as the first position vector, randomly initialize the particle swarm in the first search space of the waiting point; with the optimization objectives of minimizing the navigation path deviation and maximizing the obstacle avoidance safety distance of the waiting point, perform optimization update; Triggering the Level 1 alarm threshold triggers global multi-objective optimization: Collect the set of waiting points in the assembly layer, upper waiting layer, and lower waiting layer under the triggering of the Level 1 alarm threshold, extract the corresponding real-time UAV positions, and mark them as the second position vector. Randomly initialize the particle swarm in the second search space. With the optimization objectives of minimizing the spatial topology distribution uniformity of all waiting points, minimizing the rate of change of total cruise speed, and maximizing the obstacle avoidance safety distance, perform optimization update.

6. The high-precision obstacle avoidance navigation system for intelligent unmanned aerial vehicles based on the Internet of Things according to claim 5, characterized in that, By retrieving real-time wind field data, the updated obstacle avoidance strategy is subject to mandatory constraints, including: The association between topological nodes and links in the composite topological space is marked; where the topological node is the waiting point in the composite topological space, the link is the navigation path generated by the improved particle swarm optimization algorithm, and each link carries real-time wind field data; where the real-time wind field data includes wind speed and wind direction; Set the functional relationship between the navigation path and real-time wind field data. By obtaining the solutions of each waiting point on the navigation path generated by optimization, a wind field feature matrix is ​​formed. Perform Boolean operation on the wind field feature matrix and the standard feature matrix, filter out the wind field feature matrices that cannot meet the first rule, mark them as 0, trigger the forced constraint, immediately discard the current optimization result, and re-optimize and update.

7. The high-precision obstacle avoidance navigation system for intelligent unmanned aerial vehicles based on the Internet of Things according to claim 1, characterized in that, The obstacle avoidance strategy is modified, including: modifying the command type, modifying the content set, and modifying the relationship set; wherein, the command type is the type of execution command; the content set is the UAV landing requirements; and the relationship set is the relationship between topology nodes.

8. The high-precision obstacle avoidance navigation system for intelligent unmanned aerial vehicles based on the Internet of Things according to claim 1, characterized in that, Perform collision checks, including: Retrieve the target landing site and correlate it with the updated navigation path in time and space to establish a dynamic safety envelope; Based on the target landing site, the effective landing area is extracted. Collision verification is performed on the dynamic safety envelope within the effective landing area. Cases where the dynamic safety envelope intersects with physical obstacles are filtered out, and the corresponding landing sites are removed. After the target landing site completes the collision verification, the final landing site distribution is output and displayed.

Images