Unmanned aerial vehicle intelligent obstacle avoidance system based on laser radar

Abstract

The invention discloses an unmanned aerial vehicle intelligent obstacle avoidance system based on a laser radar, and belongs to the technical field of path planning, and the system specifically comprises a parameter collection module which is used for collecting original point cloud data of a flight area; carrying out obstacle geometric structure analysis to generate desensitization characteristic parameters including surface curvature and spatial density distribution; the feature analysis module is used for sending the desensitization feature parameters to an edge computing node through a hierarchical encryption transmission link, and performing primary aggregation on the feature parameters of a plurality of unmanned aerial vehicles in the same region to form a region feature template; the cloud server is used for integrating obstacle features of different areas through a security fusion protocol and constructing a global obstacle feature library; the path planning module is used for acquiring update data of the global obstacle feature library through the safe synchronization channel, and generating a three-dimensional obstacle avoidance path by combining the real-time scanning features with a feature library matching result; according to the invention, multi-machine collaborative obstacle avoidance path optimization based on obstacle avoidance information sharing is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of path planning, and particularly to an intelligent obstacle avoidance system for unmanned aerial vehicles based on lidar. Background Art

[0002] With the wide application of unmanned aerial vehicles (UAVs) in fields such as agricultural plant protection, power line inspection, and logistics transportation, the ability of autonomous obstacle avoidance in complex environments has become a core requirement for ensuring flight safety. Traditional obstacle avoidance systems mostly rely on lidar and vision sensors carried by a single UAV to construct a local environment map through real-time perception. However, the operation range and data acquisition duration of a single UAV are limited, making it difficult to cover the full life cycle changes of obstacles, resulting in limitations in environmental cognition in terms of locality and timeliness. In the scenario of large-scale UAV swarm operations, how to achieve multi-UAV collaborative environmental perception and knowledge sharing has become a key challenge in improving the intelligence level of the obstacle avoidance system.

[0003] In the prior art, some solutions attempt to improve the obstacle recognition ability through cloud data aggregation. For example, a centralized data processing architecture is adopted to upload the raw point cloud data of multiple UAVs to the cloud platform for unified modeling, and a deep learning algorithm is used to construct an obstacle feature library. Another solution proposes a local data sharing mechanism based on edge computing to exchange obstacle position information within the UAV formation to optimize the path planning.

[0004] In summary, the obstacle recognition models in the prior art rely on static feature matching and are difficult to adapt to the morphological changes of dynamic obstacles, resulting in certain misjudgments in the obstacle avoidance path. Moreover, directly sharing the raw point cloud data or obstacle coordinate information is likely to expose sensitive information about the operation area, posing a risk of privacy leakage. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent obstacle avoidance system for unmanned aerial vehicles based on lidar to solve the following technical problems: How to optimize the obstacle avoidance path of multi-UAV collaboration based on obstacle avoidance information sharing.

[0006] The purpose of the present invention can be achieved through the following technical solutions: An intelligent obstacle avoidance system for unmanned aerial vehicles based on lidar, comprising: A parameter acquisition module that synchronously acquires the raw point cloud data of the flight area based on lidar sensors configured on multiple UAVs; a feature processing unit local to each UAV analyzes the geometric structure of obstacles in the raw point cloud data to generate desensitized feature parameters including surface curvature and spatial density distribution; A feature analysis module for sending the desensitized feature parameters to an edge computing node through a hierarchical encryption transmission link, and the edge computing node initially aggregates the feature parameters of multiple UAVs in the same area to form a regional feature template; A cloud server for receiving regional feature templates of multiple edge computing nodes, integrating obstacle features in different regions through a secure fusion protocol, and constructing a global obstacle feature library; A path planning module for controlling each unmanned aerial vehicle to obtain updated data of the global obstacle feature library through a secure synchronization channel, and generating a three-dimensional obstacle avoidance path by combining real-time scanning features and feature library matching results.

[0007] As a further solution of the present invention: In the parameter acquisition module, the specific process of parsing the geometric structure of the obstacle is as follows: Divide the original point cloud data into three-dimensional cube units of equal volume, calculate the gradient of the change in the number of point clouds in each cube unit, and generate a spatial density distribution map; Analyze the surface continuity features between adjacent cube units, and identify rigid obstacles with regular geometric structures by calculating the direction difference value of the point cloud normal vectors between cube units; Extract the curvature mutation points at the boundaries of the cube units, construct the obstacle surface contour line, and distinguish independent obstacles and continuous obstacle groups according to the closed characteristics of the obstacle surface contour line; Fuse and encode the spatial density distribution map and the obstacle surface contour line features to generate a fixed-dimensional feature vector sequence; perform spatial position confusion processing on the feature vector sequence, and use a random matrix permutation algorithm to eliminate the geographical location correlation of the original point cloud data.

[0008] As a further solution of the present invention: In the feature analysis module, the specific working process of the hierarchical encrypted transmission link is as follows: Before each unmanned aerial vehicle sends the desensitized feature parameters, use the elliptic curve encryption algorithm to perform primary encryption on the data packet of the desensitized feature parameters, and attach an irreversible device anonymous identification code to the encrypted data packet; After the edge computing node receives the encrypted data packets of multiple unmanned aerial vehicles in the same region, use the key derivation method based on the hash chain to generate a temporary decryption key, and strip the device anonymous identification code after decryption; After the cloud server receives the regional feature templates from different edge computing nodes, use the secure multi-party computing protocol to verify the data integrity, and use the homomorphic encryption technology to perform secondary confusion on the cross-regional features; The confused feature data is attached with a time stamp and a regional code through the blind signature mechanism to generate a global feature set that cannot trace the original device source.

[0009] As a further solution of the present invention: In the cloud server, the specific process of constructing the global obstacle feature library is as follows: Build a cross - regional feature association model, compare the regional feature template with the historical feature database for similarity, and identify obstacle categories with regional particularities; Establish dynamic priority rules for recurring cross - regional obstacle features. For crop features that change over time in agricultural scenarios, assign a high update frequency weight, and for device features with fixed spatial positions in power scenarios, assign a high matching accuracy weight; Construct a three - dimensional feature projection space, map obstacle features in different regions to a unified coordinate system, form an obstacle feature relationship network covering multiple operation scenarios, establish spatio - temporal association links for obstacle features in the obstacle feature relationship network, record the morphological change rules and position migration patterns of the same type of obstacles in different seasons, and update the obstacle morphology and position according to the current time point; When the cumulative unrecognized obstacles of the drones in a certain region reach the proportion of the feature library capacity, the cloud server sends an enhanced scanning instruction to the drone group in that region, specifying the pitch angle change sequence and echo intensity acquisition parameters of the lidar; After each drone collects multi - modal feature data according to the enhanced scanning instruction, the edge computing node performs differential noise injection and spatial permutation processing on the original features; the processed feature data is uploaded to the cloud server via a security protocol, and the three - dimensional feature projection space is updated through an incremental learning algorithm; the fusion process of the new feature data and the historical data performs double verification, including feature distribution consistency verification and spatio - temporal logic continuity verification.

[0010] As a further solution of the present invention: The dynamic priority rule is specifically: Attach an environmental perception label to each regional feature template, and the environmental perception label includes the illumination condition level and surface material reflection characteristic parameters during collection; Select a feature optimization strategy according to the parameters of the environmental perception label: For obstacle features in the day - night alternation scenario, use a time - dimension sliding window optimization algorithm, and for obstacle features in the complex terrain scenario, use a space - dimension feature enhancement algorithm; Set a feature degradation coefficient in the three - dimensional feature projection space to automatically reduce the priority of historical features that exceed the survival period; Regularly scan the low - frequency feature nodes in the relationship network, and perform an archiving storage operation on the low - frequency feature nodes that have not been referenced for consecutive multiple cycles.

[0011] As a further solution of the present invention: In the path planning module, the interaction method of the secure synchronization channel is: When the cloud server sends a feature update request to the drone, attach a digitally signed regional feature digest value; After the drone receives the feature update request, compare the locally stored feature digest value with the regional feature digest value sent by the cloud, and only upload the subset of feature parameters with differences; Before receiving the updated data of the global obstacle feature library, the UAV generates a zero-knowledge proof document containing the current environmental feature parameters and submits it to the cloud server for validity verification; After the cloud server passes the verification, it distributes customized updated segments of the global obstacle feature library according to the geocoding and operation type of the area where the UAV is located; During the transmission of the updated segments, a block verification mechanism is adopted, and each data block contains a self-verifying hash value and an associated verification with adjacent data blocks.

[0012] As a further solution of the present invention: in the path planning module, the process of generating a three-dimensional obstacle avoidance path is as follows: Perform three-level matching verification on the geometric structure features extracted in real time and the global obstacle feature library: The first level matches the basic geometric contour of the obstacle and compares the surface curvature distribution with the curvature template of known obstacles in the feature library; The second level matches the spatial density features and verifies the point cloud distribution pattern and the density change law of the same type of obstacles in the feature library; The third level matches the dynamic change trend and compares the real-time scan data with the morphological evolution path of the obstacles recorded in the feature library; For obstacles that succeed in all three-level matches, call the pre-stored avoidance strategy, and start the multi-frame contour reconstruction process for obstacles that partially succeed in the match; construct a three-dimensional safe passage area according to the match verification results, and the boundary of the three-dimensional safe passage area dynamically shrinks as the obstacle recognition confidence improves.

[0013] As a further solution of the present invention: the specific process of the multi-frame contour reconstruction process is as follows: Collect multi-angle point cloud data of the same obstacle within three consecutive flight control cycles to construct a time series model of the obstacle surface contour; Analyze the volume change rate and centroid offset trajectory of the time series model of the obstacle surface contour to distinguish the motion attributes of static and dynamic obstacles; Execute a contour completion algorithm for static obstacles and perform interpolation reconstruction according to the point cloud missing area in adjacent frames; establish a motion trajectory prediction model for dynamic obstacles and calculate the spatial conflict probability between the dynamic obstacles and the UAV flight path; Upload the reconstructed complete contour and prediction trajectory to the global obstacle feature library to trigger the incremental update mechanism of the feature library.

[0014] As a further solution of the present invention: the construction process of the three-dimensional safe passage area is as follows: Divide the safety level according to the obstacle recognition confidence: set a fixed safety boundary for obstacles that are completely successfully matched, and expand the boundary of obstacles that are partially successfully matched by a preset ratio; Establish a dynamic buffer area calculation model to control the real-time flight speed and attitude adjustment ability of the drone; Fuse the overlapping parts of adjacent three-dimensional safe passage areas to generate a continuous three-dimensional flight corridor; deploy virtual navigation beacons inside the three-dimensional flight corridor, and the positions of the virtual navigation beacons are dynamically arranged according to the maximum safe distance of the surface curvature of the obstacle; when it is detected that the deviation between the virtual navigation beacon and the drone heading exceeds the threshold, trigger a local path replanning instruction.

[0015] As a further solution of the present invention: the deployment method of the virtual navigation beacon is as follows: Extract the central axis of the three-dimensional flight corridor, set the initial beacon positions at equal time intervals along the axis direction; adjust the beacon spacing according to the minimum safe diameter of the cross-section of the three-dimensional flight corridor, and the smaller the diameter, the higher the beacon density; Establish a risk monitoring area at each beacon position, and calculate the surface distance change rate between the risk monitoring area and the nearest obstacle in real time; when the surface distance change rate exceeds the safety threshold, send a position adjustment instruction to the adjacent beacon to form a new heading guidance path; The position data of the virtual navigation beacon is distributed to other drones through a secure synchronization channel to achieve collaborative obstacle avoidance path sharing.

[0016] Advantages of the present invention: Based on a hierarchical federated learning framework, the present invention constructs a global obstacle feature library. Each drone completes the geometric structure analysis and spatial position confusion processing of the point cloud data locally, generates desensitized feature parameters, and then aggregates them through elliptic curve encryption and edge nodes to achieve the secure sharing of cross-device obstacle features; through a three-dimensional feature projection space and dynamic priority rules, multi-source heterogeneous obstacle features are mapped to a unified coordinate system and a spatio-temporal association link is established. Combined with a three-level matching verification mechanism, matching analysis is carried out respectively from the basic geometric contour, spatial density characteristics, and dynamic change trend to improve the recognition accuracy of complex obstacles; adopt a multi-frame contour reconstruction process and virtual navigation beacon deployment technology, perform surface contour interpolation reconstruction and motion trajectory prediction on unmatched obstacles, construct a three-dimensional flight corridor including a dynamic buffer area, and realize the deep coupling of path planning and obstacle recognition; with the help of an incremental update trigger mechanism and a directional feature learning task, when the number of unrecognized obstacles in a specific area reaches the threshold, automatically start enhanced scanning and differential noise injection processing, and continuously optimize the spatio-temporal evolution modeling ability of the global obstacle feature library, thereby significantly reducing the misjudgment rate of new obstacles on the premise of avoiding the leakage of original data, and solving the problems of redundant obstacle avoidance paths and low operation efficiency caused by data islands in traditional solutions. Description of the Drawings

[0017] The present invention will be further described below with reference to the accompanying drawings.

[0018] Figure 1 This is a schematic diagram of the module of the present invention. Specific implementation manners

[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0020] Please refer to Figure 1 As shown, the present invention is an intelligent obstacle avoidance system for drones based on lidar, including: Parameter acquisition module: Multiple drones synchronously collect the original point cloud data of the flight area through lidar, and the local feature processing unit analyzes the geometric structure of the data. The specific process includes: dividing the point cloud into three-dimensional cube units, calculating the gradient of the number of points in the unit to generate a spatial density map; identifying rigid obstacles through the difference in normal vectors, extracting the points with curvature mutations to construct the surface contour line to distinguish independent obstacles and obstacle groups; finally, fusing the density map and contour features and encoding them into feature vectors, and eliminating the geographical location correlation through random matrix permutation to generate desensitized feature parameters.

[0021] Feature analysis module: Responsible for the secure transmission and regional aggregation of desensitized feature parameters. The drone initially encrypts the feature parameters through the elliptic curve encryption algorithm and attaches an anonymous identifier. After receiving, the edge computing node decrypts and aggregates the data in the same region through the hash chain key derivation to form a regional feature template; after receiving the template, the cloud server uses secure multi-party computing to verify the integrity, and secondarily obfuscates the features through homomorphic encryption, and finally generates a global feature set with untraceable sources.

[0022] Cloud server: Build and maintain a global obstacle feature library. Identify regionally special obstacles through a cross-regional feature association model, manage feature weights based on dynamic priority rules, such as high update frequency in agricultural scenarios and high matching accuracy in power scenarios; build a three-dimensional feature projection space, map multi-source features to a unified coordinate system, and establish a spatio-temporal association link to record the seasonal changes of the obstacle shape and position; when the number of unrecognized obstacles in the region reaches the threshold, issue an enhanced scanning instruction to continuously optimize the feature library through incremental learning.

[0023] Path planning module: Achieve incremental updates of the feature library through a secure synchronization channel. For example, by comparing digest values, only transmit differential data, and generate a 3D obstacle avoidance path in combination with real-time scan data. Specifically, it includes: verifying obstacles through three-level matching, where the matching includes geometric contours, spatial density, and dynamic trends, and initiating multi-frame contour reconstruction for partially matched obstacles; dividing safety levels according to the matching confidence, constructing a 3D flight corridor with a dynamic buffer area, and deploying virtual navigation beacons. The beacon spacing is dynamically adjusted according to the corridor width, supporting multi-aircraft collaborative path sharing.

[0024] In a preferred embodiment of the present invention, in the parameter acquisition module, the specific process of parsing the geometric structure of the obstacle is as follows: 1. Quantitative modeling of spatial density features Divide the original point cloud data collected by the lidar into equal-volume cube cells in 3D space. The cell size can be adjusted according to the operation scenario. For example, 1m³ cells are used in agricultural scenarios, and 0.5m³ cells are used in power line inspection scenarios. By calculating the gradient change of the point cloud quantity in each cell, that is, the ratio of the difference in the point cloud quantity between adjacent cells to the spatial distance, a spatial density distribution map is generated. This map visually presents the density of the spatial distribution of obstacles in the form of a heat map. For example, areas with dense point clouds show high gradient values, such as building walls; areas with sparse point clouds show low gradient values, such as low-altitude vegetation; areas without point clouds are marked as passable spaces.

[0025] 2. Identification of geometric regularity of rigid obstacles For adjacent cube cells, by calculating the direction difference value of the point cloud normal vectors between the cells, and using the cosine value of the vector angle to measure, judge the surface continuity feature. When the angle between the normal vectors is less than a preset threshold (such as 30°), it is considered that the adjacent cells belong to the same smooth surface, such as a flat wall; if the angle exceeds the threshold, it is determined as a surface mutation boundary, such as a corner or an obstacle corner. Based on this, rigid obstacles with regular geometric structures, such as cuboid buildings and cylindrical towers, can be quickly identified, and the interference of irregular obstacles such as natural terrain and dynamic vegetation can be excluded.

[0026] 3. Obstacle contour construction and cluster classification At the boundary of the cube cell, extract curvature mutation points through curvature calculation, such as the curvature value based on fitting a quadratic surface to the local point cloud. These points are usually located at the edges or vertices of obstacles, such as the edge of the tree crown and the corner of the vehicle top. Connect the curvature mutation points in sequence to form the obstacle surface contour line, and judge the obstacle type through the closure of the contour line: a closed contour line corresponds to an independent obstacle, such as a single tree or an isolated building, and a non-closed or chain-like contour line corresponds to a continuous obstacle group, such as a forest boundary or a building complex. This classification mechanism provides key topological information for subsequent path planning. For example, independent obstacles can be bypassed, and a global bypass path needs to be planned for continuous obstacle groups.

[0027] 4. Feature Fusion and Privacy Protection Processing Encode and fuse the numerical features of the spatial density distribution map (such as the gradient value of each unit) and the geometric features of the surface contour line (such as the contour line length and the coordinates of the curvature extreme points) to generate a sequence of feature vectors with a fixed dimension, such as a 128-dimensional vector, to realize the dimensionality reduction representation of obstacle features. To avoid the original data from revealing the geographical location of the operation area, a random matrix permutation algorithm is used to scramble the spatial coordinate components of the feature vectors: a random orthogonal matrix is generated to perform a linear transformation on the vectors, so that the transformed feature vectors lose their association with the original coordinate system, while retaining the relative relationships of the geometric structures, such as distance and angle invariance. This processing ensures that the desensitized feature parameters only contain the morphological features of the obstacles and do not contain any geographical location information.

[0028] In another preferred embodiment of the present invention, in the feature analysis module, the specific working process of the hierarchical encryption transmission link is as follows: 1. Initial Encryption and Anonymization at the Device End Before sending the desensitized feature parameters, the drone encrypts the data packet using the Elliptic Curve Cryptography (ECC) algorithm. The ECC algorithm is based on the elliptic curve discrete logarithm problem and has the advantages of short key length and high calculation efficiency, making it suitable for terminal devices with limited computing resources such as drones. The encrypted data packet is appended with an irreversible device anonymous identification code, which is generated by performing a hash operation (such as SHA-256) on the device physical address (MAC address), ensuring that attackers cannot trace the original device identity through the identification code. For example, the MAC address of a certain drone generates a 64-bit anonymous identification "a1b2c3..." after hashing, and this identification is only valid for the current transmission cycle and is regenerated during the next communication.

[0029] 2. Decryption, Aggregation and De-identification at the Edge Node After receiving the encrypted data packets of multiple drones, the edge computing nodes in the same area generate a temporary decryption key using the key derivation method based on the hash chain. The hash chain generates a sequence of keys through iterative hash operations, that is, H(k), H(H(k)),..., and each key is only used for decrypting the data in the current cycle to ensure the timeliness of the key. After decryption, the edge node strips the device anonymous identification code, only retains the feature parameters, and performs the initial aggregation of the feature parameters in the same area: by statistically averaging indicators such as curvature and density distribution mean, a regional feature template reflecting the regional commonality is generated, such as the average density feature of crop rows in a farmland scene. This template removes the noise data unique to a single machine and improves the robustness of the features.

[0030] 3. Secondary Confusion and Global Feature Generation in the Cloud After the cloud server receives the cross - regional area feature templates, it first verifies the data integrity through the Secure Multi - Party Computation (MPC) protocol. MPC allows the participating parties (edge nodes) to perform collaborative calculations without sharing the original data. For example, through interactive verification to ensure that the template has not been tampered with. After passing the verification, homomorphic encryption technology is used to perform secondary obfuscation on the cross - regional features: allowing numerical operations such as addition and multiplication to be directly performed on the encrypted data, enabling the cloud to aggregate features without accessing the plaintext. The obfuscated feature data is attached with a timestamp and a regional code through a blind signature mechanism. For example, "20250527 - 01" represents the data of region 1 on May 27, 2025. Finally, a global feature set that cannot trace the original device source is generated. This set only retains the cross - regional common features of obstacles, such as the transmission tower structures common in different regions, and supports the cloud to build an obstacle feature library covering multiple scenarios.

[0031] In another preferred embodiment of the present invention, in the cloud server, the specific process of constructing the global obstacle feature library is as follows: 1. Cross - regional feature association and identification of regional particularities Through a cross - regional feature association model, such as the K - Nearest Neighbor algorithm based on cosine similarity, the area feature templates uploaded by the edge nodes are compared with the historical feature database. For example, when the "crop row density feature" uploaded by a certain agricultural area has a similarity lower than a threshold (such as 60%) with the features of the same type of crops in the historical database, it is determined as an obstacle category with regional particularities (such as a new crop variety or a special planting pattern). The system automatically marks such features and triggers an artificial review process to ensure accurate modeling of rare obstacles.

[0032] 2. Dynamic priority rules: Management of scenario - sensitive features Dynamically adjust the processing strategy for obstacle features in different operation scenarios: Agricultural scenario: Crop growth is seasonal, such as sparse in the seedling stage and dense in the flowering stage. High update frequency weights are assigned to crop features that change over time (such as point cloud density, stem curvature). For example, the crop feature template is forced to be updated every 7 days to ensure that the feature library reflects the crop growth status in real - time.

[0033] Power scenario: Equipment such as transmission towers and substations has fixed positions but complex structures. High matching accuracy weights are assigned to the features of equipment with fixed spatial positions, such as the geometric profile of the tower and the curvature of the insulator. By increasing the dimension of the feature vector, such as from 128 - dimensional to 256 - dimensional, the recognition accuracy of complex structures is improved.

[0034] 3. Three - dimensional feature projection space: Unified modeling of multi - scenario features Construct a three-dimensional feature projection space. Based on the principal component analysis (PCA) or t-SNE dimensionality reduction technique, map the obstacle features in different regions and at different scales to a unified coordinate system. For example, project the features such as "crop rows" in the agricultural scenario, "towers" in the power scenario, and "buildings" in the logistics scenario into the same space to form an obstacle feature relationship network covering multiple operation scenarios. Establish spatio-temporal association links in the network to record the morphological changes of the same type of obstacles in different seasons, such as the change in point cloud density of a fruit tree from flowering to fruiting, and the position migration, such as the dynamic obstacles on the migratory path of migratory birds. The system automatically retrieves the corresponding seasonal feature template according to the current time point to update the real-time morphology and position of the obstacles.

[0035] 4. Incremental learning mechanism: Active detection of unknown obstacles When the cumulative number of unrecognized obstacles of the drones in a certain area reaches 10% (configurable threshold) of the feature library capacity, the cloud triggers an enhanced scanning instruction: Instruction parameters: Specify the pitch angle change sequence of the lidar, such as scanning once every 10° from -30° to +60°, and the echo intensity acquisition parameters, such as enhancing the signal gain of weak reflection targets, to ensure the acquisition of multi-angle and multi-intensity features of the obstacles.

[0036] Data processing: The multi-modal feature data collected by the drones is processed by the edge nodes. Through differential noise injection, that is, adding noise conforming to the Laplace distribution to satisfy differential privacy, and spatial permutation, that is, randomly shuffling the point cloud coordinate order, to further desensitize. The processed feature data is uploaded to the cloud through a secure transmission protocol, and the three-dimensional projection space is updated using an incremental learning algorithm (such as online gradient descent) to avoid the computational overhead of retraining the entire model.

[0037] Double verification: When the new features are fused with the historical data, perform the verification of the consistency of the feature distribution, such as using the K-S test to judge whether the new and old feature distributions are consistent, and the verification of the spatio-temporal logical continuity, such as whether the change in the obstacle position conforms to the geographical migration law, to ensure the reliability of the new data.

[0038] In a preferred case of this embodiment, the dynamic priority rule is specifically as follows: 1. Environmental perception label: Scene-based annotation of features Attach environmental perception labels to each regional feature template, including: Illumination condition level: Divided into three levels: low light (≤500 lux), medium light (500 - 3000 lux), and strong light (≥3000 lux), which are collected in real time by the light sensors carried by the drones.

[0039] Surface material reflection characteristics: such as soil (reflectivity 10 - 20%), vegetation (20 - 40%), metal (60 - 80%), are obtained by inverting the lidar echo intensity. The labeled data is used for the dynamic selection of subsequent feature optimization strategies.

[0040] 2. Feature optimization strategy: Adaptive processing in spatio-temporal dimensions Sliding window optimization in the time dimension (day-night scenarios): In response to the changes in obstacle shadows caused by day-night alternation, a sliding window algorithm is adopted. The window size is set to 24 hours, and only the feature data within the current window is retained, while the outdated day-night features are eliminated to improve the matching efficiency in a dynamic environment.

[0041] Feature enhancement in the spatial dimension: In terrains such as mountains and canyons, the three-dimensional structural features of obstacles (such as height differences and slopes) are highlighted through feature enhancement algorithms (such as feature vector normalization and noise filtering), and the interference of terrain noise on obstacle avoidance is suppressed.

[0042] 3. Feature degradation and archiving: Automated management of the life cycle Feature degradation coefficient: A survival period is set for each feature in the three-dimensional projection space. For example, the default is 30 days. Features that exceed the period automatically reduce the priority weight, such as decaying exponentially to 50% of the original value, to avoid the influence of outdated features on new feature matching.

[0043] Archiving of low-frequency features: Regularly scan the feature relationship network. For low-frequency features that have not been referenced by the drone for 5 consecutive cycles, such as temporary obstacles formed by rare natural disasters, perform archiving storage operations, migrating from the real-time matching library to the historical database, releasing computing resources while retaining data traceability.

[0044] In another preferred embodiment of the present invention, in the path planning module, the interaction method of the secure synchronization channel is as follows: 1. Feature update request and digest comparison Request initiated by the cloud: When the global obstacle feature library is updated, such as adding a new regional feature template or optimizing existing features, the cloud server sends a feature update request to the drone group in the target area, along with a digitally signed regional feature digest value. The digest value is generated by a hash algorithm (such as SHA-3) for the key parameters of the updated features (such as the mean of the curvature template and the variance of the density distribution), and the length is fixed at 256 bits to ensure the verifiability of data integrity.

[0045] Local Verification of UAV: After receiving a request, the UAV extracts the feature digest value stored locally, synchronizes it to the local cache regularly, and compares it bit by bit with the digest value sent from the cloud. Only when the two are inconsistent is it determined that there are feature differences, triggering the subsequent differential update process to avoid transmitting redundant data. For example, if the local digest is "d41d8cd98f00b204e9800998ecf8427e", it means there is no feature update, and the upload and download operations are skipped.

[0046] 2. Zero-Knowledge Proof: Trusted Verification of Environmental Features Before receiving the updated data from the cloud, the UAV needs to prove the legality of its current operating environment to the cloud without disclosing specific environmental parameters, which is specifically achieved through zero-knowledge proof (ZK-SNARKs) technology: Proof Generation: The UAV collects the current environmental feature parameters, such as light level, surface reflectivity, and the geocoding prefix of the real-time location, and generates a proof document containing these parameters through an encryption function. The proof document only indicates that "the current environment meets the feature update conditions", such as being within a preset operation area, but does not expose the specific parameter values.

[0047] Cloud Verification: The cloud server uses the pre-deployed verification key to verify the validity of the proof document. For example, it verifies whether the UAV is located in the authorized operation area, such as through the coordinate range verification of the geofence, or whether it belongs to a specific operation type, such as an agricultural plant protection UAV only receiving crop-related feature updates. After successful verification, the cloud unlocks the download permission for the corresponding feature update segment.

[0048] 3. Customized Feature Distribution: Scenario-Sensitive Data Push The cloud executes a differential feature update strategy based on the geocoding and operation type tags of the UAV: Geocoding Matching: Match the geocoding of the UAV's real-time location (such as the three-level coding of province-city-district based on the BD-09 coordinate system) with the area tags in the feature library (such as "North China - Beijing - Chaoyang District"), and only push the obstacle features of this area and adjacent areas (such as common local building and tree types) to reduce the transmission of irrelevant data.

[0049] Operation Type Filtering: For different operation scenarios, including agriculture, power, and logistics, distribute customized feature subsets. For example, agricultural UAVs only receive features related to crops, irrigation equipment, etc., while power UAVs receive features of equipment such as transmission towers and insulators, avoiding the redundancy of the feature library from affecting the matching efficiency.

[0050] Update Segment Compression: Losslessly compress the customized feature segments (such as using the gzip algorithm), with a compression ratio of up to 3:1, further reducing the transmission bandwidth requirements, which is suitable for UAVs operating in weak network environments in remote areas.

[0051] 4. Block Check Mechanism: Double Guarantee for Reliable Transmission The updated segments are divided into data blocks of a fixed size (such as 512 KB per block) during transmission, and double check information is attached: Self-verifying Hash Value: Each data block generates a unique hash value through the SHA-256 algorithm. After receiving, the drone recalculates the hash value and compares it with the transmitted value to ensure that a single data block has not been tampered with or damaged.

[0052] Adjacent Block Association Check: A chained check mechanism is introduced. Each data block contains the hash value of the previous data block. For example, the check field of block n contains the hash value of block n - 1, forming a check chain. If a certain data block fails to be transmitted, the error location can be quickly located through the adjacent block check, triggering single-block retransmission instead of full-scale retransmission, improving the transmission efficiency.

[0053] In another preferred embodiment of the present invention, in the path planning module, the process of generating a three-dimensional obstacle avoidance path is as follows: The generation of the three-dimensional obstacle avoidance path is based on a multi-level feature matching mechanism, improving the accuracy and robustness of obstacle recognition through progressive verification: 1. First-level Matching: Basic Geometric Profile Check The surface curvature distribution features, such as Gaussian curvature and mean curvature, are extracted from the obstacle point cloud data scanned in real time through the principal curvature calculation algorithm and compared with the curvature templates of known obstacles in the global feature library. For example: For spherical obstacles, such as water storage tanks, verify whether the surface curvature is evenly distributed and close to the theoretical spherical curvature value; For cuboid obstacles, such as buildings, check whether the edge curvature presents a right-angled mutation feature.

[0054] The matching threshold is dynamically adjusted according to the operation scenario. The power inspection scenario requires a curvature error ≤ 5%, and the agricultural scenario can be relaxed to 10% to accommodate the natural variation of crop forms.

[0055] 2. Second-level Matching: Spatial Density Feature Verification Analyze the spatial density distribution pattern of the real-time point cloud, such as the change trend of the number of point clouds with height and the geometric shape of the sparse area, and compare it with the density change law of the same type of obstacles in the feature library. For example: For low-altitude vegetation obstacles, verify whether the point cloud density presents the natural growth feature of "dense at the bottom and sparse at the top"; For transmission tower obstacles, check whether the density distribution conforms to the structural feature of "dense at the tower base and sparse at the tower frame".

[0056] Calculate the mean and variance of the density distribution through the sliding window statistical method. When the difference between the real-time data and the feature library template exceeds the preset threshold, such as the variance deviation > 15%, it is determined as a partial match.

[0057] 3. Third-level matching: Dynamic change trend analysis Retrieve the spatio-temporal evolution records of obstacles in the feature library, such as the morphological changes of the same crop in different growth cycles, the seasonal movement of the migratory path of migratory birds, and compare with the dynamic trend of the real-time scan data: For static obstacles, such as buildings, check whether their positions and forms are consistent with the historical records; For dynamic obstacles, such as moving vehicles, predict whether their future 3-second movement trajectories match the average speed and steering mode of similar objects in the feature library.

[0058] Fit the deviation between the real-time trajectory and the historical data through the Kalman filtering algorithm. When the prediction error exceeds the safe distance (such as 2 meters), it is determined that the dynamic features do not match.

[0059] 4. Disposal of matching results Full three-level matching: Directly call the pre-stored standard avoidance strategy, such as executing a fixed path of "going around the tower for half a week + maintaining a 5-meter vertical distance" for known transmission towers. Partial matching: Start the multi-frame contour reconstruction process to avoid misjudgment caused by the lack of single-frame data.

[0060] In a preferred case of this embodiment, the specific process of the multi-frame contour reconstruction process is as follows: 1. Multi-angle data acquisition and time series modeling The drone collects the point cloud data of the same obstacle within three consecutive flight control cycles (usually 0.3 seconds) at a yaw angle of ±15°, and constructs a three-dimensional contour time series model, which includes the contour coordinate sets at times t - 1, t, and t + 1. For example, for a tree blocked on the side, the complete crown contour is restored by stitching the data of the front view, left view, and right view frames.

[0061] 2. Motion attribute discrimination: Classification of static and dynamic obstacles Calculation of volume change rate: Compare the volume differences of adjacent frame contours. If the change rate < 5%, it is determined as a static obstacle, such as a fixed building, and vice versa as a dynamic obstacle, such as a moving machine.

[0062] Centroid offset analysis: Track the coordinate changes of the contour centroid, and combine the flight speed of the drone itself to calculate the relative motion speed and direction of the obstacle. For example, when the angle between the centroid offset trajectory and the drone flight path > 90°, it is determined as a lateral crossing obstacle and needs to be avoided first.

[0063] 3. Differential reconstruction and prediction algorithm Static obstacle completion: Using the triangulation interpolation algorithm, the missing areas of the point cloud in adjacent frames (such as the backlit side of buildings) are surface-fitted to complete the contour details. The contour error after completion should be < 0.2 meters (agricultural scenario) or < 0.05 meters (power scenario).

[0064] Dynamic obstacle prediction: Based on the linear regression model, a motion trajectory prediction equation is established. Combining the obstacle size and the UAV speed, the spatial conflict probability is calculated, such as the probability that the distance is less than the safety threshold within the next 5 seconds. When the probability > 60%, an emergency obstacle avoidance plan is triggered.

[0065] 4. Incremental update of the feature library The reconstructed complete contour data and the dynamic prediction model parameters (mean speed, volume change rate threshold) are uploaded to the cloud, and the global feature library is updated through the federated learning mechanism to achieve "one recognition, shared across the network". For example, for a new type of agricultural machinery obstacle that appears in a certain area for the first time, its features after reconstruction will be synchronized to all UAVs in this area to avoid repeated misjudgments.

[0066] In another preferred case of this embodiment, the construction process of the three-dimensional safe passage area is as follows: 1. Safety level division and dynamic boundary adjustment Completely matching obstacles: According to the standard size recorded in the feature library, a fixed safety boundary is delimited, such as expanding 1.5 meters around the obstacle, and the boundary remains unchanged during flight.

[0067] Partially matching / reconstructing obstacles: Expand the safety boundary by a preset ratio, such as 200%, to form a buffer safety zone to reduce the collision risk caused by incomplete contours. For example, for a reconstructed tree obstacle, the boundary is expanded from the standard 1 meter to 2 meters until the confidence level is increased to more than 80%.

[0068] 2. Coupling of the dynamic buffer area and flight control A buffer area calculation model is established, and the input parameters include: Obstacle recognition confidence level, ranging from 0 - 100%; The current speed and attitude adjustment ability of the UAV, such as the maximum turning rate; the output safety buffer distance is obtained, such as the buffer distance is 3 meters when the confidence level is 50%, and it drops to 1.5 meters when the confidence level is 80%. The model ensures the smoothness of the obstacle avoidance action by adjusting the speed limit of the UAV in real time, such as limiting the speed to 3m / s when entering the high buffer area, and the turning radius.

[0069] 3. Generation of the three-dimensional flight corridor and beacon deployment Corridor integration processing: For the overlapping part of the safety areas of adjacent obstacles, Boolean operations are used to merge them into a continuous three-dimensional flight corridor to avoid path breaks caused by discrete boundaries. For example, after the safety areas of two adjacent woods are integrated, a "passage between trees" is formed.

[0070] Dynamic arrangement of virtual navigation beacons: Set initial beacons at equal time intervals (such as the flight distance every 0.5 seconds) along the central axis of the corridor; Adjust the beacon spacing according to the minimum safety diameter of the corridor cross-section. When the diameter < 2 meters, the beacon density is doubled to ensure navigation accuracy; The beacon is built-in with a risk monitoring module to calculate the rate of change of the surface distance from the nearest obstacle in real time. When the rate of change > 0.5 m / s, an automatic position adjustment instruction is sent to the adjacent beacon to form an adaptive obstacle avoidance path.

[0071] 4. Local path replanning trigger mechanism When the deviation between the UAV heading and the beacon guiding direction exceeds the threshold, the system determines that the path is deviated and triggers local replanning: Search for an alternative path within the current corridor based on the A* algorithm, and preferentially select a channel with a small curvature change and a high beacon density; the replanning time should be < 200 ms to ensure the real-time nature of obstacle avoidance response.

[0072] It should be noted that the deployment method of the virtual navigation beacon is as follows: I. Initial layout of beacons: Axis extraction and dynamic density adjustment The deployment of virtual navigation beacons is based on a three-dimensional flight corridor, and intelligent layout is achieved through geometric feature analysis: Use a skeleton extraction algorithm to generate a central axis from the point cloud data of the three-dimensional flight corridor. This axis represents the geometric center line of the corridor to ensure that the beacons are symmetrically distributed along the path. For example, in an "L"-shaped building corridor or a curved forest path, the axis extends along the corridor direction, maintaining a balanced safety distance from the obstacles on both sides.

[0073] Set the initial beacon positions at fixed time intervals along the axis direction. For example, deploy a beacon every 0.5 seconds of the UAV's flight distance. Assuming the UAV flight speed is 6 m / s, the initial spacing between adjacent beacons is 3 meters, forming a uniformly distributed sequence of navigation nodes.

[0074] Calculate the minimum safety diameter of the flight corridor cross-section in real time. That is, in the plane perpendicular to the axis, the shortest distance between two obstacles, and dynamically adjust the beacon density according to the diameter: When the diameter ≥ 5 meters, maintain the initial spacing of 3 meters; When 3 meters ≤ diameter < 5 meters, the beacon spacing is reduced to 1.5 meters, and the density is doubled; When the diameter is less than 3 meters, further reduce the spacing to 0.8 meters to ensure dense coverage of beacons in narrow spaces, such as the gaps between transmission towers and urban alleys, and improve navigation accuracy.

[0075] II. Risk Monitoring and Beacon Dynamic Adjustment: Real-time Environment Response Mechanism Each beacon acts as an independent intelligent sensing node and has the ability to adapt to the dynamic environment: A spherical risk monitoring area is established centered on the beacon location, and the radius is dynamically adjusted according to the beacon spacing. For example, when the spacing is 1.5 meters, the radius is set to 2 meters to ensure that the monitoring range covers potential obstacles between adjacent beacons.

[0076] Continuously collect the point cloud data of obstacles in the monitoring area through lidar, and calculate the change speed of the distance between the beacon center and the surface of the nearest obstacle in real time. When the distance rapidly shrinks or expands, such as when the change rate exceeds the safety threshold of 0.6 m / s, it is determined that there is a collision risk or path redundancy.

[0077] If the distance rapidly shrinks, such as when an obstacle approaches, the beacon offsets 0.3 - 0.8 meters in the safe direction perpendicular to the axis, and sends a collaborative adjustment instruction to the adjacent beacons upstream and downstream to form a detour curve; If the distance significantly increases, such as when entering an open area, the beacon moves forward 1 - 2 meters along the axis direction to compress the redundant path and improve flight efficiency. The adjusted beacon position is quickly synchronized to adjacent nodes through a distributed communication protocol to ensure the coherence of path adjustment.

[0078] III. Cooperative Obstacle Avoidance Path Sharing: Multi-aircraft Data Distribution and Navigation Synchronization The position data of virtual navigation beacons is shared in real time within the UAV swarm through a secure synchronization channel to achieve collaborative perception of dynamic paths: The beacon position information (including three-dimensional coordinates, adjustment time, risk level) is processed by an encryption algorithm and sent to the UAV group in the same area through the multicast mode to reduce the overhead of repeated transmission. Data compression technology can reduce the transmission volume by about 50% to meet the communication requirements of UAVs in weak network environments.

[0079] After receiving the beacon data, the UAV fuses it with the point cloud of obstacles scanned locally in real time and displays it as a visual icon on the on-board interface. For example, a green beacon represents a safe path, a yellow beacon indicates the need to decelerate, and a red beacon warns of a dangerous area to assist the UAV in path prediction.

[0080] The lead UAV is responsible for detecting the forward path and generating beacons, and subsequent UAVs directly follow the beacons to fly, reducing the energy consumption of repeated scanning; When a certain drone triggers beacon adjustment due to dynamic changes in obstacles, the new beacon data is automatically synchronized to all members of the cluster, achieving the collaborative obstacle avoidance effect of "single unit perception, group response" and avoiding path conflicts caused by independent planning of multiple drones.

[0081] The above has described an embodiment of the present invention in detail, but the content described is only a preferred embodiment of the present invention and cannot be considered as limiting the scope of implementation of the present invention. All equivalent changes and improvements made according to the scope of the present invention application should still fall within the patent coverage scope of the present invention.

Claims

1. An intelligent obstacle avoidance system for drones based on lidar, characterized in that, Including: A parameter acquisition module that synchronously acquires the original point cloud data of the flight area based on lidar sensors configured on multiple drones; the feature processing units local to each drone analyze the geometric structure of obstacles in the original point cloud data to generate desensitized feature parameters including surface curvature and spatial density distribution. A feature analysis module that sends the desensitized feature parameters to an edge computing node through a hierarchical encryption transmission link, and the edge computing node initially aggregates the feature parameters of multiple drones in the same area to form a regional feature template. A cloud server that receives the regional feature templates from multiple edge computing nodes, integrates the obstacle features in different regions through a secure fusion protocol, and constructs a global obstacle feature library. A path planning module that controls each drone to obtain the updated data of the global obstacle feature library through a secure synchronization channel, and generates a three-dimensional obstacle avoidance path by combining the real-time scanning features and the matching results of the feature library.

2. The intelligent obstacle avoidance system for unmanned aerial vehicles based on lidar according to claim 1, wherein, In the parameter acquisition module, the specific process of the obstacle geometric structure analysis is as follows: Divide the original point cloud data into three-dimensional cube units of equal volume, calculate the gradient of the change in the number of point clouds in each cube unit, and generate a spatial density distribution map. Analyze the surface continuity features between adjacent cube units, and identify rigid obstacles with regular geometric structures by calculating the direction difference value of the point cloud normal vectors between cube units. Extract the curvature mutation points at the boundaries of the cube units, construct the obstacle surface contour line, and distinguish independent obstacles and continuous obstacle groups according to the closed characteristics of the obstacle surface contour line. Fuse and encode the spatial density distribution map and the obstacle surface contour line features to generate a feature vector sequence of a fixed dimension. Perform spatial position confusion processing on the feature vector sequence, and use a random matrix permutation algorithm to eliminate the geographical location correlation of the original point cloud data.

3. An intelligent obstacle avoidance system for an unmanned aerial vehicle based on lidar according to claim 1, characterized in that, In the feature analysis module, the specific working process of the hierarchical encryption transmission link is as follows: Before sending the desensitized feature parameters, each drone initially encrypts the data packet of the desensitized feature parameters using the elliptic curve encryption algorithm, and attaches an irreversible device anonymous identification code to the encrypted data packet. After receiving the encrypted data packets from multiple drones in the same area, the edge computing node generates a temporary decryption key using a key derivation method based on a hash chain, and strips the device anonymous identification code after decryption. After receiving the regional feature templates from different edge computing nodes, the cloud server uses a secure multi-party computing protocol to verify the data integrity, and uses homomorphic encryption technology to perform secondary confusion on the cross-regional features. The confused feature data is attached with a timestamp and a regional code through a blind signature mechanism to generate a global feature set that cannot trace the original device source.

4. An intelligent obstacle avoidance system for an unmanned aerial vehicle based on lidar according to claim 1, wherein, In the cloud server, the specific process of constructing the global obstacle feature library is as follows: Establish a cross-regional feature association model, compare the regional feature templates with the historical feature database for similarity, and identify obstacle categories with regional particularities. Establish dynamic priority rules for the repeated cross-regional obstacle features, assign a high update frequency weight to the crop features that change over time in the agricultural scenario, and assign a high matching accuracy weight to the device features with fixed spatial positions in the power scenario. Construct a three-dimensional feature projection space, map the obstacle features in different regions to a unified coordinate system, form an obstacle feature relationship network covering multiple operation scenarios, establish spatio-temporal correlation links for obstacle features in the obstacle feature relationship network, record the morphological change rules and position migration patterns of the same type of obstacles in different seasons, and update the obstacle morphology and position according to the current time point; When the cumulative unrecognized obstacles of the drones in a certain area reach the proportion of the feature library capacity, the cloud server sends an enhanced scanning instruction to the drone group in this area, specifying the pitch angle change sequence and echo intensity acquisition parameters of the lidar; After each drone collects multi-modal feature data according to the enhanced scanning instruction, the edge computing node performs differential noise injection and spatial permutation processing on the original features; the processed feature data is uploaded to the cloud server through a security protocol, and the three-dimensional feature projection space is updated through an incremental learning algorithm; The fusion process of the new feature data and the historical data performs double verification, including feature distribution consistency verification and spatio-temporal logic continuity verification.

5. The intelligent obstacle avoidance system for an unmanned aerial vehicle based on lidar according to claim 4, characterized in that, The specific dynamic priority rule is: Attach an environmental perception label to each regional feature template, and the environmental perception label includes the lighting condition level and surface material reflection characteristic parameters during acquisition; Select a feature optimization strategy according to the parameters of the environmental perception label: the obstacle features in the day-night alternation scenario adopt a time-dimensional sliding window optimization algorithm, and the obstacle features in the complex terrain scenario adopt a space-dimensional feature enhancement algorithm; Set a feature degradation coefficient in the three-dimensional feature projection space to automatically reduce the priority of historical features that exceed the survival period; Regularly scan the low-frequency feature nodes in the relationship network, and perform an archiving storage operation on the low-frequency feature nodes that have not been referenced for multiple consecutive cycles.

6. The intelligent obstacle avoidance system for unmanned aerial vehicle based on lidar according to claim 1, wherein, In the path planning module, the interaction method of the secure synchronization channel is: When the cloud server sends a feature update request to the drone, it attaches a region feature digest value with a digital signature; After receiving the feature update request, the drone compares the locally stored feature digest value with the region feature digest value sent by the cloud, and only uploads the subset of feature parameters with differences; Before receiving the global obstacle feature library update data, the drone generates a zero-knowledge proof document containing the current environmental feature parameters and submits it to the cloud server for validity verification; After the cloud server passes the verification, it distributes customized global obstacle feature library update segments according to the geographical coding and operation type of the area where the drone is located; A block verification mechanism is adopted during the transmission of the update segment, and each data block contains a self-verifying hash value and an associated verification with adjacent data blocks.

7. An intelligent obstacle avoidance system for an unmanned aerial vehicle based on lidar according to claim 1, characterized in that, In the path planning module, the process of generating a three-dimensional obstacle avoidance path is: Perform three-level matching verification on the geometric structure features extracted in real time and the global obstacle feature library: The first level matches the basic geometric contour of the obstacle, and compares the surface curvature distribution with the curvature template of the known obstacles in the feature library; The second level matches the spatial density feature, and verifies the point cloud distribution pattern and the density change rule of the same type of obstacles in the feature library; The third level matches the dynamic change trend, and compares the real-time scanning data with the obstacle morphological evolution path recorded in the feature library; Invoke the pre-stored avoidance strategy for obstacles with successful three-level matching, and initiate the multi-frame contour reconstruction process for obstacles with partial successful matching; construct a three-dimensional safe passage area according to the matching verification result, and the boundary of the three-dimensional safe passage area dynamically shrinks as the obstacle recognition confidence improves.

8. An intelligent obstacle avoidance system for an unmanned aerial vehicle based on lidar according to claim 7, characterized in that, The specific process of the multi-frame contour reconstruction process is as follows: Collect multi-angle point cloud data of the same obstacle within three consecutive flight control cycles to construct a time series model of the obstacle surface contour; Analyze the volume change rate and centroid offset trajectory of the time series model of the obstacle surface contour to distinguish the motion attributes of static and dynamic obstacles; Execute the contour completion algorithm for static obstacles and perform interpolation reconstruction according to the point cloud missing area of adjacent frames; establish a motion trajectory prediction model for dynamic obstacles and calculate the spatial conflict probability between the dynamic obstacle and the UAV flight path; Upload the reconstructed complete contour and predicted trajectory to the global obstacle feature library to trigger the incremental update mechanism of the feature library.

9. An intelligent obstacle avoidance system for an unmanned aerial vehicle based on lidar according to claim 7, characterized in that, The construction process of the three-dimensional safe passage area is as follows: Divide the safety level according to the obstacle recognition confidence: set a fixed safety boundary for obstacles with complete successful matching, and expand the boundary of obstacles with partial successful matching by a preset ratio; Establish a dynamic buffer area calculation model to control the real-time flight speed and attitude adjustment ability of the UAV; Fuse the overlapping parts of adjacent three-dimensional safe passage areas to generate a continuous three-dimensional flight corridor; deploy virtual navigation beacons inside the three-dimensional flight corridor, and the positions of the virtual navigation beacons are dynamically arranged according to the maximum safety distance of the obstacle surface curvature; when it is detected that the deviation between the virtual navigation beacon and the UAV heading exceeds the threshold, trigger a local path replanning instruction.

10. An intelligent obstacle avoidance system for an unmanned aerial vehicle based on lidar according to claim 9, characterized in that, The deployment method of the virtual navigation beacon is as follows: Extract the central axis of the three-dimensional flight corridor and set the initial beacon positions at equal time intervals along the axis direction; adjust the beacon spacing according to the minimum safety diameter of the cross-section of the three-dimensional flight corridor, and the smaller the diameter, the higher the beacon density; Establish a risk monitoring area at each beacon position and calculate the surface distance change rate between the risk monitoring area and the nearest obstacle in real time; when the surface distance change rate exceeds the safety threshold, send a position adjustment instruction to the adjacent beacon to form a new heading guidance path; The position data of the virtual navigation beacon is distributed to other UAVs through a secure synchronization channel to achieve collaborative obstacle avoidance path sharing.

Images