Unmanned aerial vehicle high-anti-interference obstacle avoidance system and method based on multi-sensor fusion

Abstract

The invention discloses an unmanned aerial vehicle high anti-interference obstacle avoidance system and method based on multi-sensor fusion, and the system comprises a multi-source sensing module which is used for collecting the multi-dimensional sensing data of the surrounding environment of an unmanned aerial vehicle; the front-end preprocessing and anti-interference module is connected with the multi-source sensing module, and the front-end preprocessing and anti-interference module is used for carrying out primary processing and interference suppression on the multi-dimensional sensing data to obtain processed data; the multi-sensor fusion sensing center is connected with the front-end preprocessing and anti-interference module, and the multi-sensor fusion sensing center is used for carrying out space-time alignment and hierarchical adaptive fusion on the processed data and outputting an environment situation map containing obstacle position, speed and semantic information; and the real-time obstacle avoidance planning and control module is connected with the multi-sensor fusion sensing center, and the real-time obstacle avoidance planning and control module is used for generating an anti-interference obstacle avoidance track according to the environment situation map and controlling the unmanned aerial vehicle to execute.

Description

Technical Field

[0001] This invention belongs to the field of autonomous navigation and flight safety technology for unmanned aerial vehicles (UAVs), and particularly relates to a UAV high anti-interference obstacle avoidance system and method based on multi-sensor fusion. Background Technology

[0002] Currently, mainstream drone obstacle avoidance solutions mainly rely on single or a few types of sensors: Vision-based obstacle avoidance: utilizes monocular, binocular, or multi-view cameras to perceive the environment through visual SLAM (simultaneous localization and mapping) or depth estimation. Its advantages are rich information and low cost, but it is heavily dependent on lighting conditions. Its performance drops sharply in adverse weather conditions such as low light, backlight, heavy fog, rain, and snow, or in areas with missing textures, and it also has a high computational load.

[0003] Ultrasonic obstacle avoidance is mainly used for near-field low-altitude altitude maintenance or simple obstacle avoidance. It has a short operating range, low angular resolution, and is susceptible to air turbulence and noise interference, making it unable to perform accurate obstacle contour recognition and velocity measurement.

[0004] Obstacle avoidance based on lidar: It can provide high-precision, high-resolution real-time 3D point cloud data with high perception accuracy. However, in environments with particulate matter such as rain, fog, smoke, and dust, the laser beam of traditional lidar will suffer severe scattering and attenuation, resulting in a decrease in point cloud quality; at the same time, it is expensive, and there may be mutual interference (crosstalk) between multiple similar devices.

[0005] Obstacle avoidance based on millimeter-wave radar offers good environmental adaptability, unaffected by lighting conditions or rain and fog, and can directly measure the relative velocity of targets at a moderate cost. However, its point cloud density and angular resolution are typically lower than those of lidar, limiting its ability to detect small obstacles.

[0006] In conclusion, every single sensor has its inherent physical limitations and limitations in application scenarios. In complex real-world environments, relying solely on a single sensor for obstacle avoidance decisions makes it difficult to guarantee the robustness and reliability of the system. Electromagnetic interference, severe weather, and sensor malfunctions can all lead to sensing failures and cause flight accidents.

[0007] In recent years, multi-sensor fusion technology has been regarded as an effective way to solve the above problems. However, existing fusion schemes mostly focus on simple data superposition or basic information complementarity, lacking real-time evaluation mechanisms for sensor data quality and active anti-interference design. When a sensor outputs erroneous data due to interference, simple fusion algorithms may not be able to effectively identify it, and may even make incorrect decisions due to "garbage in, garbage out," which is unacceptable in the highly dynamic and safety-critical autonomous flight of UAVs.

[0008] Therefore, there is an urgent need for a UAV high anti-interference obstacle avoidance system and method that can not only integrate information from multiple heterogeneous sensors, but also actively resist and suppress various interferences at the hardware and software levels, and make dynamic decisions based on data reliability. Summary of the Invention

[0009] To address the aforementioned technical problems, this invention proposes a high-interference-resistance obstacle avoidance system and method for unmanned aerial vehicles (UAVs) based on multi-sensor fusion, thereby resolving the issues present in the prior art.

[0010] To achieve the above objectives, the present invention provides a high-interference-resistance obstacle avoidance system for unmanned aerial vehicles based on multi-sensor fusion, comprising: Multi-source sensing module is used to collect multi-dimensional perception data of the environment around the drone; A front-end preprocessing and anti-interference module is connected to the multi-source sensing module. The front-end preprocessing and anti-interference module is used to perform preliminary processing and interference suppression on the multi-dimensional sensing data to obtain processed data. The multi-sensor fusion perception center is connected to the front-end preprocessing and anti-interference module. The multi-sensor fusion perception center is used to perform spatiotemporal alignment and hierarchical adaptive fusion on the processed data and output an environmental situation map containing obstacle position, speed and semantic information. The real-time obstacle avoidance planning and control module is connected to the multi-sensor fusion perception center. The real-time obstacle avoidance planning and control module is used to generate an anti-interference obstacle avoidance trajectory based on the environmental situation map and control the UAV to execute it.

[0011] Optionally, the multi-source sensing module includes: Anti-jamming millimeter-wave radar is used to transmit frequency-modulated continuous waves and suppress electromagnetic interference in the same frequency band to obtain the original point cloud. Anti-interference lidar is used to emit coded laser pulses and resist ambient light crosstalk to obtain the original three-dimensional point cloud; Multispectral vision unit for simultaneously acquiring visible light and infrared thermal imaging images; The auxiliary state perception unit is used to provide the pose and state information of the UAV itself.

[0012] Optionally, the front-end preprocessing and anti-interference module includes: The radar signal processing submodule is used to receive the raw point cloud output by the millimeter-wave radar and perform interference detection and spectral analysis on the millimeter-wave radar echo signal based on the raw point cloud to filter out interference. The lidar point cloud processing submodule is used to receive the original three-dimensional point cloud output by the lidar and perform empty filtering on the original three-dimensional point cloud to remove noise points. The visual image enhancement submodule is used to perform dynamic range adjustment and dehazing enhancement processing on visual images to obtain visual feature point clouds.

[0013] Optionally, the radar signal processing submodule includes: The spectrum analysis unit is used to perform short-time Fourier transform on the intermediate frequency signal of millimeter-wave radar to generate a range-Doppler spectrum; The interference identification and suppression unit is used to identify abnormal spectral peaks in the range-Doppler spectrum and suppress them using Wiener filtering to obtain a purified signal. The target detection unit is used to perform target detection and point cloud generation on the purified signal.

[0014] Optionally, the lidar point cloud processing submodule includes: An isolated point filtering unit is used to remove points whose number of neighbors is less than a threshold based on a statistical radius filter. Ground segmentation unit is used to fit the ground plane using a random sampling consensus algorithm and segment ground points from non-ground obstacle points.

[0015] Optionally, the multi-sensor fusion sensing center is constructed based on a hierarchical adaptive fusion framework, and the multi-sensor fusion sensing center includes: The geometry and dynamics layer is used to fuse the state estimation of LiDAR point cloud, visual feature point cloud and millimeter-wave radar radial velocity to output a dynamic 3D map of obstacles. The semantic and association layer is used to perform semantic recognition on visual images and map semantic labels to corresponding obstacles in the dynamic 3D obstacle map; The credibility assessment and decision-making layer is used to monitor the quality of each data stream and dynamically allocate fusion weights.

[0016] Optionally, the credibility assessment and decision-making layer includes: The health rating unit is used to calculate the radar point cloud signal-to-noise ratio, lidar point cloud density, and visual detection confidence as quality indicators in real time. The weight dynamic calculation unit is used to output the fusion weights of each data source based on the quality indicators and the changing trends of the quality indicators through a fuzzy logic controller. The final fusion output unit is used to adjust the observation noise covariance according to the fusion weights and output an environmental situation map with comprehensive confidence.

[0017] Optionally, the real-time obstacle avoidance planning and control module includes: The threat assessment unit is used to calculate threat values ​​based on obstacle distance, approach speed, and system confidence level. The model predictive control trajectory planning unit is used to construct an optimization problem that includes UAV dynamics constraints, obstacle avoidance constraints, and target approach, and to solve for the local optimal trajectory.

[0018] This invention also provides a method for high anti-interference obstacle avoidance of unmanned aerial vehicles (UAVs) based on multi-sensor fusion, applied to the aforementioned UAV high anti-interference obstacle avoidance system, comprising the following steps: Based on the synchronous acquisition of perception data of the UAV's surrounding environment and its own status by multi-source heterogeneous sensors; Perform front-end anti-interference processing on the sensed data to filter out interference signals and noise to obtain multi-source data after anti-interference processing; After anti-interference processing, the multi-source data is spatiotemporally aligned, and then hierarchical adaptive fusion is performed on the spatiotemporally aligned multi-source data to obtain an environmental situation map containing obstacle position, velocity and semantic information. Real-time threat assessment is performed based on the aforementioned environmental situation map to obtain threat assessment results; Based on the threat assessment results and fused perception information, an anti-interference obstacle avoidance trajectory is generated and the UAV is controlled to execute it.

[0019] Compared with the prior art, the present invention has the following advantages and technical effects: This invention provides a high-reliability obstacle avoidance system for unmanned aerial vehicles (UAVs) based on multi-sensor fusion. Through heterogeneous data acquisition from multi-source sensor modules, active interference suppression by front-end preprocessing and anti-interference modules, hierarchical adaptive fusion and dynamic reliability assessment by the multi-sensor fusion perception center, and threat perception trajectory generation by the real-time obstacle avoidance planning and control module, it achieves all-weather, highly reliable obstacle avoidance under complex electromagnetic environments and adverse weather conditions. The system effectively identifies and suppresses various types of sensor interference, dynamically adjusts fusion weights based on data quality, significantly improving perception robustness; and combines semantic information and motion prediction to achieve intelligent and smooth obstacle avoidance decisions, ensuring flight safety and mission continuity for UAVs in dynamic environments. Attached Figure Description

[0020] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a system structure diagram of an embodiment of the present invention. Detailed Implementation

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0022] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0023] Example 1 This invention aims to address the core technical problems of existing UAV obstacle avoidance systems, such as unreliable perception, simplistic decision-making, and poor fault tolerance in complex, dynamic, and interference-prone environments. To this end, this invention proposes a systematic solution that goes beyond simply combining multiple sensors; it constructs an adaptive system that deeply integrates active anti-interference mechanisms and intelligent information processing from the physical layer to the decision-making layer.

[0024] like Figure 1 As shown, this embodiment provides a complete obstacle avoidance system implementation plan mounted on a hexacopter UAV platform. The core computing unit of the system uses a high-performance embedded processor, responsible for running all algorithm modules except for the underlying sensor drivers. Each hardware module is connected to the computing unit via a high-speed bus, and global time consistency of data acquisition is ensured through a hardware time synchronization signal.

[0025] This embodiment provides a high anti-interference obstacle avoidance system for unmanned aerial vehicles based on multi-sensor fusion, including: a multi-source sensing module, a front-end preprocessing and anti-interference module, a multi-sensor fusion perception center, and a real-time obstacle avoidance planning and control module.

[0026] The hardware selection and configuration of the multi-source sensing module are as follows: Anti-jamming millimeter-wave radar: A frequency-modulated continuous wave radar module with a center frequency of 77 GHz was selected. Its built-in RF front-end integrates an adaptive interference suppression unit based on a field-programmable gate array. This unit performs the following process in real time: performing a short-time Fourier transform on the intermediate frequency signal after analog-to-digital conversion to generate a range-Doppler spectrum; in the spectrum, identifying spectral peaks with energy significantly higher than the background noise through constant false alarm rate detection; combining prior knowledge, zeroing out the spectral peak regions suspected of interference or suppressing them using Wiener filtering; finally, performing an inverse transform to obtain the purified signal, and then performing conventional target detection and point cloud generation.

[0027] Anti-interference lidar: A 16-line mechanical rotating lidar is used, and its transmission drive circuit is modified to emit a set of pseudo-random binary sequence code modulated laser pulses. The photoelectric signal at the receiving end is amplified and sent to a dedicated matched filter processor. This processor stores a copy of the same pseudo-random code, and through correlation calculations, it greatly improves the signal-to-noise ratio of the real echo signal.

[0028] Multispectral vision unit: Employs a global shutter industrial RGB camera and an infrared thermal imager with an uncooled vanadium oxide focal plane array. The two are rigidly connected and jointly calibrated to ensure their optical axes are approximately parallel, with known pixel-level correspondences.

[0029] Auxiliary Status Awareness Unit: Employs a combined navigation module integrating a 9-axis IMU and a dual-frequency GNSS receiver, as well as a high-precision digital barometer.

[0030] Front-end preprocessing and anti-interference module: This module runs in parallel on the computing unit using a multi-threaded approach. Specifically, it includes: Radar signal processing sub-thread: Receives the raw point cloud output from the millimeter-wave radar. First, preliminary screening is performed based on the signal-to-noise ratio and spatial clustering characteristics of the point cloud. Next, for point sets belonging to the same suspected object, secondary verification is performed using the consistency of their Doppler velocities to filter out false points caused by multipath effects.

[0031] The LiDAR point cloud processing sub-thread receives the raw 3D point cloud output from the encoded LiDAR. First, a statistical radius filter is applied to remove isolated points whose number of points in a given radius neighborhood is less than a threshold. Then, a random sample consensus algorithm is used to fit the ground plane, segmenting the point cloud into ground points and non-ground obstacle points.

[0032] The visual image processing sub-thread processes RGB images using a fast dehazing algorithm based on dark channel priors and applies gamma correction to adjust image contrast. For infrared images, it performs non-uniformity correction and dynamic range stretching. The processed images, along with pose information from the auxiliary state perception unit, are used to calculate visual feature points and their 3D positions.

[0033] Multi-sensor fusion sensing center: The fusion sensing center is the core of the algorithm, and its three-level processing flow is as follows: Level 1: Fusion of geometry and dynamics layers. This layer maintains a global state vector X, which contains the UAV's own state and a dynamically managed list of obstacle states.

[0034] Prediction steps: Predict the state vector X based on IMU data and a high-frequency motion model.

[0035] Update steps: This is a multi-observation-source update process: LiDAR observation: The segmented non-ground point cloud clusters of the current frame are associated with the predicted obstacle boundaries in the state vector. Successfully matched point cloud clusters are used to update the position and boundary dimensions of the corresponding obstacles. Unmatched point cloud clusters may represent new obstacles and will be initialized.

[0036] Millimeter-wave radar observation: Correlating the radar point cloud with obstacles in the state vector. Once the correlation is successful, the precise radial velocity provided by the radar will be used as a strong constraint to update the velocity vector of the obstacles.

[0037] Visual geometric observation: 3D feature points extracted from the image are associated with obstacles or environmental structures in the state vector to fine-tune the position estimation.

[0038] The output of this layer is a real-time, dynamic list of obstacles with their position, velocity, and physical boundaries.

[0039] Level 2: Semantic and Relational Layer Fusion. This layer runs a lightweight convolutional neural network to perform simultaneous inference on the preprocessed RGB and infrared images, obtaining the bounding boxes of obstacles in the images, their semantic categories, and confidence scores.

[0040] Projection association: Using the camera's intrinsic and extrinsic parameters, the 3D bounding box of each obstacle output from the first stage is projected onto the image plane to form a 2D projection box.

[0041] Data Association: Calculate the intersection-union ratio (IoU) between each projected bounding box and the 2D bounding boxes detected by deep learning. Matching pairs with an IoU greater than a threshold and a reasonable semantic category are considered successfully associated. At this point, the detected semantic labels and high-level features are assigned to the corresponding obstacle states at the first level.

[0042] Level 3: Credibility Assessment and Decision-Making Integration. This level maintains a real-time "health" score H_i(t) for each data source and the integrated output of Levels 1 and 2.

[0043] Metrics calculation: For radar, health is related to the average signal-to-noise ratio and clustering stability of its point cloud; for lidar, it is related to point cloud density and ground fitting residual; for visual semantics, it is related to the average confidence score of the detection box.

[0044] Dynamic weight calculation: A simplified fuzzy logic controller is used. The input is the current H_i(t) of each data source and its changing trend, and the output is the fusion weight W_i(t) used for final state estimation.

[0045] Final Fusion and Output: In the update step, the first-level state estimator multiplies the noise covariance matrix of each observation by the inverse of its corresponding data source weight. Finally, the system outputs an environmental situation map, where each obstacle is accompanied by: high-precision 3D position and velocity, physical dimensions, semantic category, and a system-level confidence C that integrates all sources.

[0046] A real-time obstacle avoidance planning and control module, connected to the multi-sensor fusion perception center, is used to generate an anti-interference obstacle avoidance trajectory based on the environmental situation map and control the UAV to execute it. The specific implementation process of the real-time obstacle avoidance planning and control module includes: 1. Threat Assessment: For each obstacle in the aforementioned environmental situation map... Calculate its instantaneous threat value The calculation formula is as follows: in: Indicates drones and obstacles The minimum Euclidean distance to the boundary; Indicates drones and obstacles The approach velocity scalar (a positive value indicates approach). This indicates the output of the fusion perception center regarding obstacles. System-level overall confidence level of the state; This is the gain coefficient adjusted based on the semantic category of the obstacle.

[0047] 2. Model-based predictive control trajectory planning: System Model: The UAV is simplified into a dual integrator model with acceleration limits in both the horizontal and vertical directions.

[0048] Optimize problem formulation: State variables: future The planned location sequence of drones at each time step ( ).

[0049] constraint: Dynamic constraints: , ,in These are the planned speed and acceleration, respectively.

[0050] Obstacle constraints: for each prediction time step and each obstacle ,Require: in It is an obstacle exist Predicted location at time and These are the equivalent radii of the drone and the obstacle, respectively. Is it related to confidence level? Additional safety margin with negative correlation.

[0051] Boundary constraints: .

[0052] objective function : Among them: the first item Drive the drone to approach the local target point ; Second item The penalty is excessive acceleration to ensure a smooth trajectory; the third item Penalty for collisions with all obstacles, The penalty function is based on distance (e.g., the penalty value increases sharply when the distance is less than a safety threshold), and its weights are... Can be compared with threat value Related.

[0053] As a specific implementation of this embodiment, dynamic obstacle avoidance in urban logistics delivery involves a drone flying between buildings to deliver goods. An RGB camera identifies "pedestrians" and "vehicles" ahead, a LiDAR precisely outlines their contours, and a millimeter-wave radar accurately detects the vehicle's lateral movement. A fusion center assigns high weights to various data types and outputs a high-confidence situation map. An MPC planner predicts the vehicle's trajectory, generating a smooth arc that bypasses the vehicle from behind, successfully avoiding obstacles.

[0054] As a specific implementation of this embodiment, robust perception under adverse weather conditions is demonstrated: A UAV performs power line inspection in light fog. Visual image quality degrades, and semantic confidence decreases. The lidar point cloud becomes sparse due to fog particles. Millimeter-wave radar performance remains largely unaffected. The third-layer fusion decision layer automatically reduces the weights of visual and lidar data while increasing the weight of millimeter-wave radar. Although the output semantic information may be temporarily reduced, radar-based geometric and motion perception remains robust. The MPC planner employs a conservative planning approach with an increased safety margin to ensure safe flight of the UAV.

[0055] Example 2 The core of the technical solution of this invention consists of two mutually supportive parts: a highly integrated hardware system and an intelligent processing method running on it.

[0056] In terms of system design, this invention presents a modular UAV high-interference-resistance obstacle avoidance system. The system uses a "multi-sensor fusion perception center" as its computational brain, whose inputs are connected to a specially designed "multi-source sensing module" and a "front-end preprocessing and anti-interference module," while its outputs drive a "real-time obstacle avoidance planning and control module." The system comprises the following components: a multi-source sensing module for collecting multi-dimensional perception data of the UAV's surrounding environment; a front-end preprocessing and anti-interference module connected to the multi-source sensing module for preliminary processing and interference suppression of the perception data; a multi-sensor fusion perception center connected to the front-end preprocessing and anti-interference module for receiving the processed data, performing spatiotemporal alignment and hierarchical adaptive fusion, and outputting an environmental situation map containing obstacle position, velocity, and semantic information; and a real-time obstacle avoidance planning and control module connected to the multi-sensor fusion perception center for generating an anti-interference obstacle avoidance trajectory based on the environmental situation map and controlling the UAV to execute it. The multi-source sensing module includes at least a detection unit with active anti-interference capabilities, comprising a frequency-modulated continuous wave millimeter-wave radar and an anti-interference lidar. The hierarchical adaptive fusion framework employed by the multi-sensor fusion perception center includes at least a data association layer for geometric and velocity information fusion, and a credibility assessment and decision layer for dynamically adjusting the weights of each sensor's data. The hierarchical adaptive fusion framework also includes a mid-level semantic fusion layer for associating obstacle semantic categories obtained based on visual image recognition with objects in the enhanced environmental map generated by the data association layer.

[0057] The multi-source sensing module is not a simple stack of sensors, but a carefully selected and configured heterogeneous array: it includes an anti-interference millimeter-wave radar that uses a frequency-modulated continuous wave system and integrates adaptive filtering circuits to effectively suppress electromagnetic interference in the same frequency band; it includes an anti-interference lidar that uses unique coded transmission and matched reception technology to resist ambient light and crosstalk between devices; it also includes a multispectral vision unit (such as a global shutter RGB camera and an infrared thermal imaging camera) to obtain rich texture and thermal radiation information; and an auxiliary state perception unit (IMU, GNSS, etc.) that provides a body state reference.

[0058] The front-end preprocessing and anti-interference module provides the first "firewall" and enhancement processing for various sensor data. For example, it performs real-time interference detection and spectral analysis on millimeter-wave radar echoes; it applies spatiotemporal filtering to lidar point clouds to remove rain and fog noise; and it performs dynamic range adjustment and dehazing enhancement on visual images. The quality and reliability of the data processed by this module are initially improved.

[0059] The front-end preprocessing and anti-interference module includes: a radar signal processing submodule, used for interference detection and filtering of millimeter-wave radar echoes; a lidar point cloud purification submodule, used for removing noise points through spatiotemporal filtering; and a visual image enhancement submodule, used for improving the quality of visual images.

[0060] The core innovation of this invention lies in the "three-level hierarchical adaptive fusion framework" adopted by the multi-sensor fusion sensing center. This framework is a dynamic, self-evaluating closed-loop processing flow: The first level (geometry and dynamics layer) focuses on the accurate reconstruction and motion estimation of the physical world. It utilizes state estimation methods such as extended Kalman filters or factor graph optimization to tightly couple the precise point cloud from LiDAR, the feature point cloud from vision, and the radial velocity directly measured by millimeter-wave radar. Its output is a real-time, dynamic 3D occupancy grid map or target list containing obstacle positions and velocity vectors, providing an accurate geometric and kinematic basis for obstacle avoidance decisions.

[0061] The second level (semantic and association layer) aims to endow the physical world with the ability to "understand." It utilizes lightweight deep learning models to perform real-time semantic segmentation or object detection on RGB and infrared images, identifying obstacle categories (such as vehicles, pedestrians, trees, and power lines). Subsequently, through coordinate transformation and data association algorithms, these semantic labels are accurately mapped to corresponding objects in the dynamic map generated in the first level, forming a semantic scene map. This enables the system to distinguish obstacles of different types.

[0062] The third level (Confidence Assessment and Decision Layer) is the "central nervous system" and fault-tolerant core of the entire perception system. It continuously monitors the quality metrics of all input data streams and intermediate outputs from the first two levels (such as signal-to-noise ratio, point cloud density, target detection confidence, and data consistency). Based on a set of fuzzy logic rules or a lightweight neural network model, this layer dynamically calculates and assigns real-time weights to each data source in the final fusion decision. For example, when the system detects that a camera image is overexposed and its confidence has decreased due to strong backlighting, it automatically reduces the weight of visual data while increasing the weight of millimeter-wave radar and lidar data that are unaffected by lighting conditions. In extreme cases (such as sensor failure), this layer can trigger system reconfiguration, shielding the faulty channel and ensuring that the system can still operate safely in degraded mode based on the remaining reliable sensors.

[0063] In terms of methodology, this invention provides a closed-loop obstacle avoidance method corresponding to the aforementioned system. The method begins with system initialization and multi-source data synchronous acquisition, followed by a parallel front-end anti-interference processing flow. After spatiotemporal alignment, the processed data is fed into the aforementioned three-level hierarchical adaptive fusion framework for iterative processing, ultimately outputting a unified environmental situation map with spatiotemporal labels, semantic information, and confidence assessment. Based on this high-dimensional situation map, the method further performs threat assessment, comprehensively considering distance, approach speed, obstacle type, and perception confidence to calculate real-time risk. Finally, Model Predictive Control (MPC) is used as the trajectory planner, incorporating UAV dynamic constraints, predicted trajectories of dynamic obstacles, and perception uncertainties into the optimization problem. This allows for the online solution of a locally optimal trajectory that is safe (away from all threats), smooth (consistent with flight characteristics), and mission-oriented over a future period, forming a high-frequency closed loop of "perception-planning-control."

[0064] In terms of perception capabilities, this invention achieves robust all-weather perception across various weather and lighting conditions. Regarding system reliability, through intrinsic credibility assessment and dynamic weight adjustment, it possesses excellent anti-interference and fault tolerance capabilities, achieving functional safety. In terms of decision-making intelligence, it provides a richer environmental cognition dimension than traditional methods, supporting more refined and human-like obstacle avoidance strategies. In terms of dynamic performance, by combining precise speed perception and MPC forward planning, it can gracefully and smoothly avoid moving obstacles. This system and method are particularly suitable for advanced UAV applications with extremely high requirements for safety, reliability, and autonomy, such as logistics delivery, power line inspection, and urban air traffic.

[0065] Example 3 This embodiment also provides a high anti-interference and avoidance method for UAVs based on multi-sensor fusion, including the following steps: S1: Multi-source heterogeneous data of the environment are collected synchronously through the multi-source sensing module, and anti-interference processing is performed through the front-end preprocessing and anti-interference module; S2: At the multi-sensor fusion perception center, the processed data is spatiotemporally aligned and hierarchical adaptive fusion is performed to generate a unified environmental situation map containing multi-dimensional information about obstacles. S3: Perform real-time threat assessment based on the aforementioned environmental situation map; S4: Perform anti-interference and fault-tolerant logic judgment, dynamically evaluate the data quality of each sensor, and adaptively adjust the sensor weights in the fusion strategy; S5: Based on the threat assessment results and the adjusted fusion information, the real-time obstacle avoidance planning and control module generates a robust obstacle avoidance trajectory and controls the UAV to execute it.

[0066] The hierarchical adaptive fusion described in step S2 includes at least the following: through a data association layer, the velocity measurement information of the millimeter-wave radar is tightly coupled with the high-precision point cloud of the lidar to generate an enhanced 3D environment map. Through credibility assessment and decision-making, dynamic weights are assigned to each sensor's information based on real-time calculated data quality indicators.

[0067] The anti-interference fault-tolerant logic judgment in step S4 specifically includes: monitoring at least one quality indicator among the signal-to-noise ratio and confidence score of each sensor data stream; when the quality indicator of any sensor is continuously lower than a preset threshold, it is determined that it is being interfered with, and its weight is reduced in the fusion decision.

[0068] When generating the obstacle avoidance trajectory in step S5, the position of static obstacles, the predicted trajectory of dynamic obstacles, and the perception uncertainty caused by sensor interference are all taken into account.

[0069] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations 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. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A high-interference-resistance obstacle avoidance system for unmanned aerial vehicles (UAVs) based on multi-sensor fusion, characterized in that, include: Multi-source sensing module is used to collect multi-dimensional perception data of the environment around the drone; A front-end preprocessing and anti-interference module is connected to the multi-source sensing module. The front-end preprocessing and anti-interference module is used to perform preliminary processing and interference suppression on the multi-dimensional sensing data to obtain processed data. The multi-sensor fusion perception center is connected to the front-end preprocessing and anti-interference module. The multi-sensor fusion perception center is used to perform spatiotemporal alignment and hierarchical adaptive fusion on the processed data and output an environmental situation map containing obstacle position, speed and semantic information. The real-time obstacle avoidance planning and control module is connected to the multi-sensor fusion perception center. The real-time obstacle avoidance planning and control module is used to generate an anti-interference obstacle avoidance trajectory based on the environmental situation map and control the UAV to execute it.

2. The UAV high anti-interference obstacle avoidance system based on multi-sensor fusion according to claim 1, characterized in that, The multi-source sensing module includes: Anti-jamming millimeter-wave radar is used to transmit frequency-modulated continuous waves and suppress electromagnetic interference in the same frequency band to obtain the original point cloud. Anti-interference lidar is used to emit coded laser pulses and resist ambient light crosstalk to obtain the original three-dimensional point cloud; Multispectral vision unit for simultaneously acquiring visible light and infrared thermal imaging images; The auxiliary state perception unit is used to provide the pose and state information of the UAV itself.

3. The UAV high anti-interference obstacle avoidance system based on multi-sensor fusion according to claim 1, characterized in that, The front-end preprocessing and anti-interference module includes: The radar signal processing submodule is used to receive the raw point cloud output by the millimeter-wave radar and perform interference detection and spectral analysis on the millimeter-wave radar echo signal based on the raw point cloud to filter out interference. The lidar point cloud processing submodule is used to receive the original three-dimensional point cloud output by the lidar and perform empty filtering on the original three-dimensional point cloud to remove noise points. The visual image enhancement submodule is used to perform dynamic range adjustment and dehazing enhancement processing on visual images to obtain visual feature point clouds.

4. The UAV high anti-interference obstacle avoidance system based on multi-sensor fusion according to claim 3, characterized in that, The radar signal processing submodule includes: The spectrum analysis unit is used to perform short-time Fourier transform on the intermediate frequency signal of millimeter-wave radar to generate a range-Doppler spectrum; The interference identification and suppression unit is used to identify abnormal spectral peaks in the range-Doppler spectrum and suppress them using Wiener filtering to obtain a purified signal. The target detection unit is used to perform target detection and point cloud generation on the purified signal.

5. The UAV high anti-interference obstacle avoidance system based on multi-sensor fusion according to claim 3, characterized in that, The lidar point cloud processing submodule includes: An isolated point filtering unit is used to remove points whose number of neighbors is less than a threshold based on a statistical radius filter. Ground segmentation unit is used to fit the ground plane using a random sampling consensus algorithm and segment ground points from non-ground obstacle points.

6. The UAV high anti-interference obstacle avoidance system based on multi-sensor fusion according to claim 1, characterized in that, The multi-sensor fusion sensing center is constructed based on a hierarchical adaptive fusion framework, and the multi-sensor fusion sensing center includes: The geometry and dynamics layer is used to fuse the state estimation of LiDAR point cloud, visual feature point cloud and millimeter-wave radar radial velocity to output a dynamic 3D map of obstacles. The semantic and association layer is used to perform semantic recognition on visual images and map semantic labels to corresponding obstacles in the dynamic 3D obstacle map; The credibility assessment and decision-making layer is used to monitor the quality of each data stream and dynamically allocate fusion weights.

7. The UAV high anti-interference obstacle avoidance system based on multi-sensor fusion according to claim 6, characterized in that, The credibility assessment and decision-making layer includes: The health rating unit is used to calculate the radar point cloud signal-to-noise ratio, lidar point cloud density, and visual detection confidence as quality indicators in real time. The weight dynamic calculation unit is used to output the fusion weights of each data source based on the quality indicators and the changing trends of the quality indicators through a fuzzy logic controller. The final fusion output unit is used to adjust the observation noise covariance according to the fusion weights and output an environmental situation map with comprehensive confidence.

8. The UAV high anti-interference obstacle avoidance system based on multi-sensor fusion according to claim 7, characterized in that, The real-time obstacle avoidance planning and control module includes: The threat assessment unit is used to calculate threat values ​​based on obstacle distance, approach speed, and system confidence level. The model predictive control trajectory planning unit is used to construct an optimization problem that includes UAV dynamics constraints, obstacle avoidance constraints, and target approach, and to solve for the local optimal trajectory.

9. A method for high anti-interference obstacle avoidance of unmanned aerial vehicles (UAVs) based on multi-sensor fusion, applied to a UAV high anti-interference obstacle avoidance system as described in any one of claims 1-8, characterized in that, Includes the following steps: Based on the synchronous acquisition of perception data of the UAV's surrounding environment and its own status by multi-source heterogeneous sensors; Perform front-end anti-interference processing on the sensed data to filter out interference signals and noise to obtain multi-source data after anti-interference processing; After anti-interference processing, the multi-source data is spatiotemporally aligned, and then hierarchical adaptive fusion is performed on the spatiotemporally aligned multi-source data to obtain an environmental situation map containing obstacle position, velocity and semantic information. Real-time threat assessment is performed based on the aforementioned environmental situation map to obtain threat assessment results; Based on the threat assessment results and fused perception information, an anti-interference obstacle avoidance trajectory is generated and the UAV is controlled to execute it.

Images