An unmanned aerial vehicle intelligent obstacle avoidance method and system based on multi-sensor fusion

Abstract

The present application relates to the technical field of intelligent obstacle avoidance, and particularly relates to a UAV intelligent obstacle avoidance method and system based on multi-sensor fusion, which comprises: collecting environmental data by a UAV and dividing the space into sectors; calculating the radial distance standard deviation and average echo intensity of the laser radar data in each sector, constructing a sensor effectiveness measurement model based on medium scattering characteristics, and obtaining an effectiveness index; using D-S evidence theory to combine the effectiveness index to adaptively correct the original obstacle confidence, and obtaining the trust degree of the corrected obstacle; fusing the trust degree of the corrected obstacle with the trust degree provided by the visual sensor, and executing a hierarchical obstacle avoidance strategy according to the final fusion confidence. The present application can effectively suppress false alarms in a water mist environment while ensuring safety at close range, and improves the operation efficiency of the UAV in complex agricultural and forestry environments.

Description

Technical Field

[0001] This invention relates to the field of intelligent obstacle avoidance technology, and in particular to an intelligent obstacle avoidance method and system for unmanned aerial vehicles (UAVs) based on multi-sensor fusion. Background Technology

[0002] With the rapid development of the low-altitude economy, drones are increasingly used in agricultural plant protection, forestry patrols, and power line inspections. Especially in unstructured environments such as mountainous and hilly areas, drone operations face enormous challenges in environmental perception. These areas are often dotted with tea gardens and woodlands, with complex terrain, and are frequently accompanied by morning fog, mountain mists, or high-concentration water mist from pesticide spraying, making the environmental medium extremely complex.

[0003] Current drone obstacle avoidance systems primarily rely on lidar and visual sensors. LiDAR measures distance by emitting lasers and receiving echoes, while visual sensors extract depth through image features. However, in actual operations, water droplets cause strong diffuse reflection of the laser, resulting in lidar receiving a large number of cluttered, short-range echoes. Existing data processing techniques typically rely on a single distance threshold for judgment, easily misinterpreting these false echoes caused by water droplets as nearby tree branches or obstacles. This forces drones to frequently perform unnecessary emergency hovering or detours in open areas, severely reducing operational efficiency.

[0004] Furthermore, most existing multi-sensor fusion algorithms (such as traditional Kalman filtering or weighted averaging) employ fixed sensor weight strategies. In environments with dynamically changing fog concentrations, the reliability of LiDAR and visual sensors fluctuates drastically. Fixed-weight fusion logic cannot adapt to this sensor performance drift caused by changes in the environmental medium, easily introducing noisy data and leading to obstacle avoidance decision errors. Therefore, there is an urgent need for a data processing method that can dynamically evaluate the effectiveness of sensor data based on the scattering characteristics of the environmental medium and adaptively adjust the fusion strategy. Summary of the Invention

[0005] To address the technical problems of frequent false alarms under water mist interference and the inability of fixed-weight fusion strategies to adapt to dynamic environmental changes, this invention provides an intelligent obstacle avoidance method and system for unmanned aerial vehicles (UAVs) based on multi-sensor fusion.

[0006] In a first aspect, the present invention provides an intelligent obstacle avoidance method for unmanned aerial vehicles (UAVs) based on multi-sensor fusion, employing the following technical solution: A method for intelligent obstacle avoidance in unmanned aerial vehicles (UAVs) based on multi-sensor fusion includes the following steps: The UAV is equipped with lidar and binocular vision sensors to collect environmental data in real time during flight, establish a body coordinate system to complete the spatial registration of heterogeneous data, and divide the forward detection field of view of the UAV into multiple independent detection sectors in the horizontal direction; acquire the point cloud set in each detection sector, calculate the radial distance standard deviation and average echo intensity of the point cloud set, construct an effectiveness measurement model based on the medium scattering characteristics, and calculate the effectiveness index of lidar data in the current detection sector; The distance safety margin factor is calculated based on the average measurement distance of the current detection sector and the preset emergency braking distance. Using the effectiveness index and the distance safety margin factor, the initial confidence level calculated based on the original distance measurement value is corrected by an adaptive correction rule to obtain the corrected confidence level of the obstacle. The corrected confidence level of the obstacle is synthesized with the confidence level provided by the binocular vision sensor using the DS evidence synthesis rule to obtain the final fusion confidence level. A graded obstacle avoidance strategy is then executed based on the comparison result between the fusion confidence level and the preset threshold.

[0007] By constructing an effectiveness measurement model using the radial distance standard deviation and average echo intensity of the point cloud set within the detection sector, it is possible to accurately distinguish between water mist interference and real physical obstacles from the perspective of physical statistical characteristics. This solves the problem of frequent false alarms in water mist environments caused by existing technologies relying solely on a single distance threshold. Simultaneously, by combining a distance safety margin factor to adaptively correct the trust level and using DS evidence synthesis rules to fuse visual information, dynamic weight adjustment under different environmental medium concentrations is achieved. This not only avoids false landings caused by thin fog but also ensures safety through a fallback mechanism at extremely close distances, improving the obstacle avoidance robustness and operational efficiency of UAVs in complex unstructured environments.

[0008] The preferred expression for the effectiveness index is:

[0009] In the formula, Indicates the first The effectiveness index of lidar data within a detection sector; Indicates the first The standard deviation of radial distance of the point cloud set within a probe sector; Indicates the first The average echo intensity of the point cloud set within each detection sector; It is the strength reference constant; The feature sensitivity coefficient; To determine the critical threshold.

[0010] By employing a specific variant of the Sigmoid function as the effectiveness index expression, and utilizing the ratio of radial distance standard deviation to average echo intensity, physical scattering characteristics are mapped to probability values ​​between 0 and 1. This allows the effectiveness index to rapidly approach 0 when water mist is detected and approach 1 when an entity is detected, thereby achieving a quantitative assessment of the degree of interference from environmental media and providing a precise mathematical basis for subsequent adaptive fusion strategies.

[0011] Preferably, the feature sensitivity coefficient A value greater than 0 is used to balance the dimensional differences between the radial distance standard deviation and the average echo intensity; the value of the intensity reference constant is set to the system noise floor intensity level of the sensor.

[0012] By setting a feature sensitivity coefficient, the dimensional differences between the radial distance standard deviation and the average echo intensity are effectively balanced, and the intensity reference constant is set to the system noise floor level to prevent the denominator from being zero. This ensures the numerical stability and computational rationality of the algorithm when processing weak echoes or extreme data, and improves the engineering applicability of the system.

[0013] The preferred expression for the modified trust level of the obstacle is:

[0014] In the formula, Indicates the corrected number The level of trust in the existence of obstacles in each sector; Indicates the first The initial confidence level for each sector is calculated based on the original ranging values; Indicates the first The effectiveness index of lidar data within a detection sector; Indicates the first Distance safety margin factor for lidar data within each detection sector.

[0015] At normal distances, the confidence level is dynamically adjusted based on the effectiveness of the lidar data, which greatly suppresses interference signals caused by water mist. Even at extremely close distances, when the effectiveness index is low, the confidence level of obstacles can be forcibly increased through the distance safety margin factor, which reflects the safety principle of "better to have a false alarm than a collision" and effectively prevents the risk of missed detection in dense fog at close range.

[0016] Preferably, the distance safety margin factor is calculated as follows: the preset emergency braking distance is used as the numerator, the sum of the average measurement distance of the corresponding detection sector and a very small positive number to prevent the denominator from being zero is used as the denominator, and the ratio of the numerator to the denominator is used as the distance safety margin factor.

[0017] The ratio of the preset emergency braking distance to the current measured distance is defined as the safety margin factor. When the distance to the obstacle approaches the physical braking limit of the drone, the value of this factor increases sharply, thereby dominating the calculation result of the trust level in the subsequent correction steps. This ensures that the drone has an absolute priority obstacle avoidance response capability within the dangerous distance, thus guaranteeing flight safety.

[0018] Preferably, the initial confidence level is calculated as follows: calculate the difference between the warning distance and the average value of the lidar data, divide the difference by the warning distance to obtain a ratio, and take the maximum value between the ratio and zero as the initial confidence level.

[0019] A normalized initial confidence calculation method based on warning distance is provided, which can transform the raw ranging readings of lidar into a standard probabilistic form, facilitating subsequent weighted calculations with the effectiveness index and unified-dimensional multi-source information fusion with other heterogeneous sensor data.

[0020] Preferably, the graded obstacle avoidance strategy based on the comparison result of the fusion confidence score and the preset threshold includes: when the final fusion confidence score is greater than 0.7, it is determined to be a real physical obstacle, and the drone is controlled to perform emergency hovering or path replanning; when the final fusion confidence score is less than or equal to 0.7, it is determined to be a safe environment or traversable thin fog, and the drone is controlled to maintain the current flight path and continue operation.

[0021] A clear fusion confidence threshold is set to implement a graded obstacle avoidance strategy. For physical obstacles with high confidence, emergency hovering or replanning is performed, while for thin fog or safe environments with low confidence, the flight path is maintained and operations continue. This minimizes the interruption of operations caused by environmental interference while ensuring flight safety and improves the smoothness of automated operation of UAVs.

[0022] Preferably, environmental data includes lidar data and visual data.

[0023] By leveraging the precise ranging capabilities of lidar and the texture and clarity perception capabilities of visual sensors to complement each other, a rich and multi-dimensional environmental perception data foundation is provided for multi-sensor fusion algorithms, overcoming the perception limitations of a single sensor in specific environments.

[0024] Preferably, the confidence level provided by the visual sensor is the image sharpness, which is obtained by processing the acquired image with the Laplacian operator and calculating the variance of its response.

[0025] By using the Laplacian operator to process images and calculate variance to obtain the confidence level of visual sensors, the reliability of the visual system in the current environment can be effectively judged. As an independent source of evidence in the DS evidence theory, it further enhances the adaptability of the final fusion result to environmental changes.

[0026] Secondly, the present invention provides an intelligent obstacle avoidance system for unmanned aerial vehicles (UAVs) based on multi-sensor fusion, employing the following technical solution: A multi-sensor fusion-based intelligent obstacle avoidance system for unmanned aerial vehicles (UAVs) includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the aforementioned multi-sensor fusion-based intelligent obstacle avoidance method for UAVs.

[0027] The aforementioned method for intelligent obstacle avoidance of unmanned aerial vehicles based on multi-sensor fusion is generated into a computer program and stored in a memory for loading and execution by a processor. Thus, a system is created based on the memory and processor for convenient use.

[0028] The present invention has the following technical effects: To address the issue of false alarms caused by water mist interference in lidar, an effectiveness measurement model based on radial distance standard deviation and echo intensity was constructed to accurately distinguish between water mist and physical objects. Combined with a distance safety fallback mechanism, obstacle confidence is dynamically adjusted, and visual information is fused using DS evidence theory. This method can suppress false alarms caused by water mist at long distances, improving operational efficiency, and can also forcibly increase safety weights at close distances to prevent missed detections, achieving a balance between anti-interference and flight safety in complex environments. Attached Figure Description

[0029] Figure 1 This is a comparison chart of the original ranging data monitored by the UAV sensor of this invention.

[0030] Figure 2 This is a graph showing the effectiveness evaluation of the sensor based on the scattering characteristics of the medium according to the present invention.

[0031] Figure 3 This is a comparison chart of the probability output of obstacle presence in this invention.

[0032] Figure 4 This is a flowchart of an intelligent obstacle avoidance method for unmanned aerial vehicles based on multi-sensor fusion according to the present invention. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] This invention discloses an intelligent obstacle avoidance method for unmanned aerial vehicles (UAVs) based on multi-sensor fusion, referring to... Figure 4 As shown, it includes the following steps: The S101 drone is equipped with lidar and visual sensors to collect environmental data, establishes a body coordinate system to complete the spatial registration of heterogeneous data, and divides the drone's forward detection field of view into multiple independent detection sectors in the horizontal direction.

[0035] In one embodiment, the UAV is equipped with a lidar and a binocular vision sensor to collect environmental data in real time during flight. To achieve spatial alignment of heterogeneous data, a UAV body coordinate system is first established. Using a pre-calibrated rotation and translation matrix, the 3D point cloud data from the lidar is projected onto the 2D image plane of the vision sensor, completing spatial registration. Subsequently, the forward detection field of view of the UAV is divided into N independent detection sectors in the horizontal direction, for example, into 10 sectors, each covering a certain angular range. For each sector i, the system synchronously acquires the point cloud set within that time window. Thus, through spatial registration and sector division, spatial alignment of heterogeneous sensor data was achieved, laying a spatial foundation for subsequent feature analysis and data fusion targeting local areas.

[0036] S102, acquire the point cloud set in each detection sector, calculate the radial distance standard deviation and average echo intensity of the point cloud set, construct a sensor effectiveness measurement model based on medium scattering characteristics, and calculate the effectiveness index of the lidar data in the current detection sector.

[0037] In one embodiment, this step aims to address the problem of distinguishing between water mist echoes and physical echoes. Physical laws indicate that physical obstacles such as tea trees and utility poles have relatively flat surfaces and high material density, resulting in concentrated distance distribution and high echo intensity for their laser echo characteristics; while water mist, being a loose medium composed of suspended particles, exhibits a chaotic distance distribution and significant echo intensity attenuation. Based on this principle, a formula for calculating the effectiveness index is constructed:

[0038] In the formula, Indicates the first The validity index of lidar data within a detection sector, with a value range of (0,1); Indicates the first The standard deviation of radial distance of point cloud sets within each sector, in meters (m); Indicates the first The average echo intensity of the point cloud set within a sector, in intensity units; This is the intensity reference constant, which represents the system noise floor level of the sensor. For example, its value is 1.0 to prevent the denominator from being zero. The characteristic sensitivity coefficient is used to balance the dimensions of the numerator and denominator and control the slope of the Sigmoid function. Its unit is intensity units per meter, and its value can be set according to the environmental scattering characteristics. To determine the critical threshold, the center offset of the classification is controlled. The value is set manually according to the actual situation; for example, the value is 5.

[0039] For example, setting , , .

[0040] Scenario 1 (Physical Obstacles): Tea trees are detected in a certain sector, with a concentrated distance distribution. Meters, with high strength. Then the fraction in the exponent term is , Close to 0, therefore The value approaches 1, indicating that the probability of the data coming from a physical obstacle is extremely high, and the data is valid.

[0041] Scenario 2 (Water Mist Interference): Morning fog exists in a certain sector, causing laser diffuse reflection and resulting in drastic distance jumps. Meters, strength degradation is severe. The exponential part is To enhance the suppression effect on strongly scattering media such as water mist, the characteristic sensitivity coefficient α is set to a relatively large value, for example, 20. , This indicates that the data most likely originated from water mist and has low validity.

[0042] In summary, The closer the value is to 1, the more the echo matches the characteristics of an entity; the closer it is to 0, the more it matches the characteristics of water mist. By constructing an effectiveness index based on the ratio of dispersion to intensity, the credibility of the current data can be accurately assessed from a statistical characteristic perspective, effectively distinguishing water mist interference from real obstacles, and providing a reliable basis for subsequent fusion decisions.

[0043] S103, calculate the distance safety margin factor based on the average measurement distance of the current detection sector and the preset emergency braking distance. Using the effectiveness index and the distance safety margin factor, correct the initial trust level calculated based on the original distance measurement value through an adaptive correction rule to obtain the corrected trust level of the obstacle.

[0044] In one embodiment, to ensure safety, inefficient data cannot be simply discarded, as even readings taken at extremely close range may pose a collision risk, even in fog. Therefore, a distance safety margin factor is introduced. The expression is:

[0045] In the formula, Indicates the first Distance safety margin factor for lidar data within each detection sector The preset emergency braking distance, for example, 2 meters. Indicates the first Average measurement distance within each detection sector, It is a very small positive number to prevent the denominator from being zero; for example, its value is 0.001.

[0046] The revised expression for the trust level of obstacles is:

[0047] In the formula, This indicates the level of confidence that the i-th sector has an obstacle after the correction, and its value ranges from [0,1]. This represents the initial confidence level of the i-th sector calculated based on the original ranging value; Indicates the first The effectiveness index of lidar data within a detection sector; Indicates the first Distance safety margin factor for lidar data within each detection sector.

[0048] in, The expression is: , The preset warning distance, for example, 10 meters. For the first The average value of lidar data within each detection sector.

[0049] The formula includes a feature confidence term and a distance catch-all term. The feature confidence term is the first term, and the distance catch-all term is the second term. The feature confidence term is derived from the original confidence level. With the effectiveness index Direct multiplication yields the result when Preserve original trust when high. Distance-based fallback option targets... In lower cases, introduce Compensation will be provided.

[0050] For example, setting The initial ranging method indicated an obstacle. rice.

[0051] Scenario A (long-distance water mist): It was determined to be fog. Rice, then The first term (feature confidence term): Second item (distance safety net): ; The trust level dropped significantly from 0.9 to 0.225, the system suppressed false alarms, and the drone continued to fly.

[0052] Scenario B (Very close range water mist / obstacle): Even ,but Rice. Then First item (feature confidence): Second item (distance safety net): ; The trust level was forcibly increased to 0.738 by the distance factor, reflecting the safety principle of "better to give a false alarm than to collide at extremely close range".

[0053] Thus, by introducing an adaptive correction mechanism with a distance safety margin, interference signals are significantly suppressed at long distances to improve efficiency, while safety weights are automatically increased at close distances to prevent missed detections, thus balancing operational efficiency and flight safety.

[0054] S104, using the DS evidence synthesis rule, the corrected level of trust in the presence of obstacles is synthesized with the level of trust provided by the visual sensor to obtain the final fusion confidence. Then, a graded obstacle avoidance strategy is executed based on the comparison result between the fusion confidence and a preset threshold.

[0055] In one embodiment, the DS evidence synthesis rule is used to adjust the degree of trust in the existence of obstacles. The confidence level is orthogonally combined with the confidence level provided by the visual sensor to obtain the final fusion confidence level. The confidence level provided by the visual sensor is image sharpness, obtained by processing the acquired image using the Laplacian operator and calculating the variance of the response result; this is existing technology. A tiered obstacle avoidance strategy is then implemented based on this confidence level. like If the obstacle is identified as a real physical obstacle, an emergency hovering or path replanning will be executed. like If the environment is determined to be safe or traversable thin fog, the drone will maintain its current flight path and continue operating.

[0056] refer to Figure 1 In the complex environment of tea gardens, the comparison between the measured distance and the actual physical distance by lidar is as follows: Figure 1 As shown in the figure, the measured values ​​of the lidar are compared with the actual physical distances, and a horizontal safety warning distance line is marked. Within the water mist interference zone, a large number of false near-range noise caused by the water mist appeared, leading to drastic changes in the measured values. Only within the actual obstacle zone did the measured values ​​remain consistent with the actual physical distances. This indicates that distance measurements alone cannot distinguish between near-range water mist and near-range objects, resulting in frequent false alarms.

[0057] refer to Figure 2This invention describes the change of the lidar effectiveness index over time. In water mist interference zones, the algorithm utilizes the extremely high data dispersion to calculate a very low effectiveness index, allowing the system to accurately classify it as non-physical interference. In real obstacle zones, the data is stable, resulting in a high effectiveness index, which the system uses to definitively identify them as physical obstacles. This demonstrates that the feature analysis algorithm of this invention can accurately assess the degree of interference from the environmental medium on the sensor.

[0058] refer to Figure 3 The final obstacle confidence output of the present invention was compared with that of existing technologies. The red obstacle avoidance trigger threshold line is shown in the figure. In the water mist area, existing technologies caused continuous false alarms due to excessive confidence in the water mist area; in contrast, the present invention achieved accurate suppression, maintaining a low confidence level, allowing the drone to pass through normally. In the real obstacle area, the confidence level of the present invention rapidly increased and stably exceeded the threshold line, achieving accurate obstacle avoidance triggering and ensuring flight safety.

[0059] Thus, by synthesizing and classifying multi-source evidence, the robustness of the system to environmental judgment is further improved, the problem of false alarms in water mist environment is solved, and a balance between anti-interference and security is achieved.

[0060] This invention also discloses an intelligent obstacle avoidance system for unmanned aerial vehicles (UAVs) based on multi-sensor fusion, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement an intelligent obstacle avoidance method for UAVs based on multi-sensor fusion according to the present invention.

[0061] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0062] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent obstacle avoidance of unmanned aerial vehicles (UAVs) based on multi-sensor fusion, characterized in that, Including the following steps: The UAV is equipped with lidar and binocular vision sensors to collect environmental data in real time during flight, establish a body coordinate system to complete the spatial registration of heterogeneous data, and divide the forward detection field of view of the UAV into multiple independent detection sectors in the horizontal direction; acquire the point cloud set in each detection sector, calculate the radial distance standard deviation and average echo intensity of the point cloud set, construct an effectiveness measurement model based on the medium scattering characteristics, and calculate the effectiveness index of lidar data in the current detection sector; Effectiveness index of lidar data within a detection sector for: Indicates the first The standard deviation of radial distance of the point cloud set within a probe sector; Indicates the first The average echo intensity of the point cloud set within each detection sector; It is the strength reference constant; The feature sensitivity coefficient; To determine the critical threshold; The distance safety margin factor is calculated based on the average measurement distance of the current detection sector and the preset emergency braking distance. Using the effectiveness index and the distance safety margin factor, the initial confidence level calculated based on the original distance measurement value is corrected through an adaptive correction rule to obtain the corrected confidence level of the obstacle. Using the DS evidence synthesis rule, the corrected obstacle confidence level is synthesized with the confidence level provided by the binocular vision sensor to obtain the final fusion confidence level. A graded obstacle avoidance strategy is then executed based on the comparison result between the fusion confidence level and a preset threshold. Trust level regarding obstacles in each sector for: Indicates the first The initial confidence level for each sector is calculated based on the original ranging values; Indicates the first The effectiveness index of lidar data within a detection sector; Indicates the first Distance safety margin factor for lidar data within each detection sector.

2. The intelligent obstacle avoidance method for unmanned aerial vehicles based on multi-sensor fusion according to claim 1, characterized in that, Feature sensitivity coefficient A value greater than 0 is used to balance the dimensional differences between the radial distance standard deviation and the average echo intensity; the value of the intensity reference constant is set to the system noise floor intensity level of the sensor.

3. The intelligent obstacle avoidance method for unmanned aerial vehicles based on multi-sensor fusion according to claim 1, characterized in that, The distance safety margin factor is calculated as follows: the preset emergency braking distance is used as the numerator, the sum of the average measurement distance of the corresponding detection sector and a very small positive number to prevent the denominator from being zero is used as the denominator, and the ratio of the numerator to the denominator is used as the distance safety margin factor.

4. The intelligent obstacle avoidance method for unmanned aerial vehicles based on multi-sensor fusion according to claim 3, characterized in that, The initial confidence level is calculated as follows: calculate the difference between the warning distance and the average value of the lidar data, divide the difference by the warning distance to obtain a ratio, and take the maximum value between the ratio and zero as the initial confidence level.

5. The intelligent obstacle avoidance method for unmanned aerial vehicles based on multi-sensor fusion according to claim 1, characterized in that, The graded obstacle avoidance strategy is implemented based on the comparison results between the fusion confidence score and the preset threshold. When the final fusion confidence score is greater than 0.7, it is determined to be a real physical obstacle, and the drone is controlled to perform emergency hovering or path replanning. When the final fusion confidence level is less than or equal to 0.7, it is determined to be a safe environment or traversable thin fog, and the drone is controlled to maintain the current flight path and continue operation.

6. The intelligent obstacle avoidance method for unmanned aerial vehicles based on multi-sensor fusion according to claim 1, characterized in that, Environmental data includes LiDAR data and visual data.

7. The intelligent obstacle avoidance method for unmanned aerial vehicles based on multi-sensor fusion according to claim 1, characterized in that, The confidence level provided by the visual sensor is the image sharpness, which is obtained by processing the acquired image with the Laplacian operator and calculating the variance of its response.

8. A drone intelligent obstacle avoidance system based on multi-sensor fusion, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a UAV intelligent obstacle avoidance method based on multi-sensor fusion according to any one of claims 1-7.

Images