Adaptive obstacle avoidance method and system for low-altitude aircraft based on ai fusion perception

Abstract

The application provides an AI fusion perception-based low-altitude aircraft adaptive obstacle avoidance method and system. The method comprises the following steps: acquiring perception data collected by each perception sensor on a target low-altitude aircraft, and performing scene recognition based on the perception data to obtain a current flight scene label; performing sensor type division on each perception sensor based on the current flight scene label to determine a decision-level fusion sensor and a data-level fusion sensor; performing obstacle detection on the perception data of each decision-level fusion sensor respectively, and fusing each detection result to obtain a first fusion detection result; fusing the perception data of each data-level fusion sensor to generate fusion data, and obtaining a second fusion detection result according to the fusion data; and realizing autonomous obstacle avoidance path planning of the target low-altitude aircraft based on the first fusion detection result and the second fusion detection result. The application can realize high-precision and high-reliability autonomous obstacle avoidance of the low-altitude aircraft.

Description

Technical Field

[0001] This invention relates to the field of multi-source data fusion technology, and in particular to an adaptive obstacle avoidance method and system for low-altitude aircraft based on AI-based fusion perception. Background Technology

[0002] With the widespread adoption and application of low-altitude aircraft in civilian and industrial scenarios such as power line inspection, urban low-altitude patrol, low-altitude logistics delivery, and field operations, autonomous obstacle avoidance capability in complex low-altitude environments has become a core technology for ensuring flight safety and improving operational reliability. Accurate and stable obstacle perception is a prerequisite for autonomous obstacle avoidance. Currently, the industry generally adopts multi-sensor cross-modal perception fusion solutions to balance detection performance under different complex scenarios. Among these, the decision-level post-fusion architecture is widely used in low-altitude aircraft obstacle detection systems due to its simple structure and strong real-time performance.

[0003] The current mainstream decision-level fusion technology approach is as follows: relying on sensors such as visible light cameras, infrared cameras, lidar, and millimeter-wave radar to independently complete obstacle detection, and then performing unified voting or weighted fusion of the top-level obstacle detection results and target confidence and other high-level semantic information output by each sensor, and then performing autonomous obstacle avoidance path planning based on the fused obstacle information.

[0004] However, existing technologies often employ fixed, undifferentiated global fusion strategies or optimize fusion results through simple dynamic weight adjustments. This approach has several drawbacks. First, erroneous results from sensors failing in harsh environments directly participate in the global fusion voting, continuously polluting the final result and easily leading to false obstacle detections, severely reducing the safety and robustness of obstacle avoidance in aircraft flight. Second, relying solely on top-level obstacle detection results for cross-modal matching and fusion results in insufficient feature representation capabilities and significant cross-modal matching difficulties for small, weak-feature obstacles such as power lines, steel wires, and thin ropes, leading to frequent missed detections. Consequently, the environmental adaptability advantages of each sensor cannot be fully utilized, resulting in low cross-modal information utilization. This makes it difficult to simultaneously meet the dual requirements of accurate detection of small obstacles and resistance to interference in harsh environments, failing to satisfy the current requirements for high-precision, high-reliability autonomous obstacle avoidance operations for low-altitude aircraft. Summary of the Invention

[0005] This invention provides an adaptive obstacle avoidance method and system for low-altitude aircraft based on AI fusion perception, in order to solve the problem that existing fusion perception technologies are unable to support low-altitude aircraft to perform accurate and reliable autonomous obstacle avoidance.

[0006] In a first aspect, embodiments of the present invention provide an adaptive obstacle avoidance method for low-altitude aircraft based on AI-fusion perception, comprising: Acquire the perception data collected by each perception sensor on the target low-altitude aircraft, and perform scene recognition based on the perception data to obtain the current flight scene label; Based on the current flight scenario label, the sensor types of each of the aforementioned sensing sensors are classified to determine the decision-level fusion sensor and the data-level fusion sensor. Obstacle detection is performed on the perception data of each of the decision-level fusion sensors, and the detection results are fused to obtain the first fusion detection result; The sensed data from each of the data-level fusion sensors are fused to generate fused data, and a second fusion detection result is obtained based on the fused data; Based on the first fusion detection result and the second fusion detection result, the autonomous obstacle avoidance path planning of the target low-altitude aircraft is realized.

[0007] In one possible implementation, scene recognition is performed based on the aforementioned perception data to obtain a current flight scene label, including: Environmental statistical features of each of the aforementioned sensing sensors are extracted based on the sensing data. Scene identification is performed based on the environmental statistical characteristics to obtain the current flight scene label.

[0008] In one possible implementation, scene recognition is performed based on the environmental statistical features to obtain the current flight scene label, including: Based on the environmental statistical features, scene category identification is performed to obtain the current scene category label; Based on the current major scene labels, select the effective sensors among the various perception sensors, and perform obstacle detection on the perception data of each effective sensor. Based on the detection results, perform subdivided scene recognition to obtain the current flight scene label.

[0009] In one possible implementation, scene recognition is performed based on the environmental statistical features to obtain the current flight scene label, including: Based on the environmental statistical features, scene recognition is performed to obtain a first flight scene label; Obstacle detection is performed on each of the aforementioned perception data, and the structured features of each detection result are extracted. Scene recognition is then performed based on the structured features to obtain a second flight scene label. The current flight scenario label is obtained based on the first flight scenario label and the second flight scenario marker.

[0010] In one possible implementation, the step of classifying each of the sensing sensors based on the current flight scenario label to determine the decision-level fusion sensor and the data-level fusion sensor includes: Based on the current flight scenario label and the preset scenario and sensor adaptation rule base, the basic scenario adaptation score of each of the sensing sensors is obtained; Based on the sensing data, the environmental statistical features of each sensing sensor are extracted, and the dynamic correction score of each sensing sensor is calculated based on the environmental statistical features. The corresponding basic scene adaptation score is corrected based on the dynamic correction score of each of the sensing sensors to obtain the modal effective score of each of the sensing sensors. The effective scores of each modality are compared with the decision-level fusion admission score threshold and the data-level fusion admission score threshold, respectively. Based on the comparison results, the sensor types of each sensing sensor are classified to determine the decision-level fusion sensor and the data-level fusion sensor.

[0011] In one possible implementation, the autonomous obstacle avoidance path planning of the target low-altitude aircraft is achieved based on the first fused detection result and the second fused detection result, including: Target association matching is performed on the obstacles in the first fusion detection result and the obstacles in the second fusion detection result to obtain the target association matching result; For the target obstacles that are successfully matched in the target association matching results, a first confidence weight and a second confidence weight of the second fusion detection results are determined based on the current flight scene label, and the information of the target obstacles is fused according to the first confidence weight and the second confidence weight; Based on the information of other obstacles besides the target obstacle in the target association matching result and the fused target obstacle information, the autonomous obstacle avoidance path planning of the target low-altitude aircraft is realized.

[0012] Secondly, embodiments of the present invention provide an adaptive obstacle avoidance system for low-altitude aircraft based on AI-fusion perception, comprising: The first processing module is used to acquire the perception data collected by each perception sensor on the target low-altitude aircraft, and to perform scene recognition based on the perception data to obtain the current flight scene label. The second processing module is used to classify the sensor types of each of the perception sensors based on the current flight scenario label, and to determine the decision-level fusion sensor and the data-level fusion sensor. The first fusion module is used to perform obstacle detection on the perception data of each of the decision-level fusion sensors, and fuse the detection results to obtain the first fusion detection result; The second fusion module is used to fuse the sensing data of each of the data-level fusion sensors to generate fused data, and to obtain a second fusion detection result based on the fused data; An adaptive obstacle avoidance module is used to realize autonomous obstacle avoidance path planning for the target low-altitude aircraft based on the first fusion detection result and the second fusion detection result.

[0013] Thirdly, embodiments of the present invention provide a low-altitude aircraft based on AI fusion perception, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect or any possible implementation of the first aspect.

[0014] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect or any possible implementation thereof.

[0015] Fifthly, embodiments of the present invention provide a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect or any possible implementation thereof.

[0016] This invention provides an AI-based fusion perception adaptive obstacle avoidance method and system for low-altitude aircraft. The method first acquires perception data collected by various sensors on the target low-altitude aircraft and performs scene recognition based on this data to obtain a current flight scene label. Then, based on the current flight scene label, the sensor types of each sensor are classified, identifying decision-level fusion sensors and data-level fusion sensors. Obstacle detection is then performed on the perception data from each decision-level fusion sensor, and the detection results are fused to obtain a first fusion detection result. The perception data from each data-level fusion sensor are then fused to generate fusion data, and a second fusion detection result is obtained based on this fusion data. Finally, autonomous obstacle avoidance path planning for the target low-altitude aircraft is implemented based on both the first and second fusion detection results. By identifying sensors with high reliability under the current flight scenario label through scene recognition, decision-level fusion sensors can be selected to achieve rapid perception of macroscopic obstacles. At the same time, sensors adapted to the current environment (i.e., corresponding to the current flight scenario label) can be selected as data-level fusion sensors to fuse raw perception data, thereby fully mining the weak features of small obstacles. The feature representation capability is improved by fusing the detection results of the two paths, reducing the probability of false detection or missed detection, and thus balancing the real-time performance and detection accuracy of obstacle detection. This provides support for achieving high-precision and high-reliability autonomous obstacle avoidance for low-altitude aircraft. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the implementation of the AI-based fusion perception adaptive obstacle avoidance method for low-altitude aircraft provided in this embodiment of the invention. Figure 2 This is a schematic diagram of the structure of the AI-based fusion perception adaptive obstacle avoidance system for low-altitude aircraft provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a low-altitude aircraft based on AI fusion perception provided in an embodiment of the present invention. Detailed Implementation

[0019] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, specific embodiments will be described below in conjunction with the accompanying drawings.

[0021] See Figure 1 The document illustrates a flowchart of the implementation of the AI-based fusion perception adaptive obstacle avoidance method for low-altitude aircraft provided in this embodiment of the invention, detailed below: In step 101, the perception data collected by each perception sensor on the target low-altitude aircraft is acquired, and scene recognition is performed based on the perception data to obtain the current flight scene label.

[0022] For example, low-altitude aircraft can be drones, electric vertical takeoff and landing (eVTOL) aircraft, etc., such as drones for power line inspection, low-altitude logistics and cargo drones, low-altitude security patrol drones, marine or riverside surveying and monitoring drones, and urban low-altitude manned eVTOL aircraft. The sensing sensors on the target low-altitude aircraft can include visible light cameras, infrared cameras, monocular vision sensors, binocular vision sensors, multi-view vision sensors, lidar, millimeter-wave radar, etc. The specific sensing sensors configured can be determined based on the usage scenario.

[0023] In this embodiment, considering that during the process of autonomous obstacle avoidance by low-altitude aircraft based on the perception data collected by various perception sensors, due to the complexity of the operating environment of low-altitude aircraft, some perception sensors may fail in some scenarios, thereby affecting the reliability and accuracy of autonomous obstacle avoidance by low-altitude aircraft. Therefore, it is considered to first perform scene recognition based on the perception data collected by various perception sensors to obtain the current flight scene label of the target low-altitude aircraft.

[0024] Optionally, scene recognition based on various sensing data to obtain the current flight scene label may include: extracting environmental statistical features of each sensing sensor based on various sensing data; and performing scene recognition based on the environmental statistical features to obtain the current flight scene label.

[0025] In this embodiment, considering the computational limitations of low-altitude aircraft, when performing scene recognition based on various sensing data, only environmental statistical features are extracted from each sensing data. These include, for example, the average brightness of the visible light camera image, global fog coefficient, image signal-to-noise ratio, and proportion of reflective pixels; the image temperature difference distribution, overall thermal imaging signal-to-noise ratio, and ambient temperature range of the infrared camera image; the average reflection intensity of the point cloud, effective point cloud density, and long-distance point cloud attenuation rate of the lidar image; and the echo signal-to-noise ratio, rain and fog attenuation factor, and velocity signal effectiveness of the millimeter-wave radar image. Then, all environmental statistical features can be input into a lightweight scene classifier to complete global scene determination and output a unique main scene label (i.e., the current flight scene label).

[0026] For example, flight scene labels may include, but are not limited to, open airspace on clear days, urban areas with strong backlighting, rain / fog / haze environments, dark environments at night, power lines with dense cables, outdoor airspace with dense forests, and riverside airspace with reflective water surfaces. This embodiment does not limit the specific flight scene labels.

[0027] For example, when outputting flight scene labels, the scene confidence level corresponding to the flight scene labels can also be output simultaneously, so as to make accurate subsequent processing using the flight scene labels.

[0028] In one embodiment, scene recognition based on environmental statistical features to obtain the current flight scene label may include: Scene categories are identified based on environmental statistical characteristics to obtain the current scene category label.

[0029] Based on the current major scene labels, the effective sensors among the various perception sensors are selected, and obstacle detection is performed on the perception data of each effective sensor. Based on the detection results, the scene is further subdivided and identified to obtain the current flight scene label.

[0030] In this embodiment, based on scene recognition using environmental statistical features, a two-stage progressive scene determination method combining coarse screening and fine judgment is adopted to improve recognition accuracy while considering computational power consumption. The first stage, coarse screening, relies on the environmental statistical features of the raw sensor data to quickly classify basic environmental categories (such as strong light, rain / fog, nighttime, and clear weather), rapidly eliminating obviously faulty sensors. The second stage, fine recognition, inputs the detection results from the effective sensors after coarse screening into a fine classifier to identify and subdivide the operational scene (such as areas with dense cabling, densely built-up areas, and open fields). Thus, the current flight scene label is accurately obtained by combining the results of the two stages of scene recognition.

[0031] In the scene recognition method of this embodiment, the low-altitude aircraft can dynamically select the scene recognition level according to the working mode. For example, in the high-speed cruise mode, only the first level of coarse recognition is run, and the second level of fine recognition is started only in the close-range high-risk airspace, so as to dynamically control the consumption of computing power.

[0032] In one embodiment, scene recognition based on environmental statistical features to obtain the current flight scene label may include: Scene identification is performed based on environmental statistical characteristics to obtain the first flight scene label.

[0033] Obstacle detection is performed on each sensing data, and the structured features of each detection result are extracted. Scene recognition is then performed based on the structured features to obtain a second flight scene label.

[0034] Obtain the current flight scenario label based on the first flight scenario label and the second flight scenario marker.

[0035] In this embodiment, to ensure the accuracy of scene recognition while taking into account computational overhead, a two-path lightweight feature parallel input scene classifier is constructed. Lightweight branch 1 does not perform complete inference of the image or point cloud, but only extracts one-dimensional environmental statistics such as average image brightness, point cloud signal-to-noise ratio, and millimeter-wave echo noise value. Lightweight branch 2 only extracts the structured features of the detection results of each sensor, such as the number of obstacles, average confidence, and target size distribution. The first flight scene label and the second flight scene label obtained by the dual features can be mutually verified. An anomaly in a single branch will not directly lead to the complete failure of scene recognition, thus providing stronger fault tolerance.

[0036] In step 102, the sensor types of each perception sensor are classified based on the current flight scenario label, and decision-level fusion sensors and data-level fusion sensors are determined.

[0037] Optionally, step 102 may include: Based on the current flight scenario labels and the preset scenario and sensor adaptation rule library, the basic scenario adaptation score of each perception sensor is obtained.

[0038] Environmental statistical features of each sensing sensor are extracted based on the various sensing data, and dynamic correction scores of each sensing sensor are calculated based on these environmental statistical features.

[0039] The corresponding basic scene adaptation score is corrected based on the dynamic correction score of each sensing sensor to obtain the effective modal score of each sensing sensor.

[0040] The effective scores of each modality are compared with the decision-level fusion admission score threshold and the data-level fusion admission score threshold, respectively. Based on the comparison results, the sensor types of each sensing sensor are classified to determine the decision-level fusion sensor and the data-level fusion sensor.

[0041] In this embodiment, a preset scene and sensor adaptation rule base can be obtained through experiments or data statistics. This rule base can contain the basic scene adaptation score for each sensor in different scenarios. For example, in rain and fog scenarios, the basic scene adaptation score for millimeter-wave radar is 90, and for visible light cameras, it is 30. Then, real-time sensing data is used to correct the basic scene adaptation score. For instance, if the global fog coefficient extracted from the real-time sensing data of the visible light camera is too high, 20 points can be deducted from the basic scene adaptation score to obtain the effective modal score for the visible light camera.

[0042] After obtaining the modal effective score for each sensing sensor using the above method, two independent admission score thresholds can be pre-configured, corresponding to the decision-level fusion branch and the data-level fusion branch, respectively. Each modal effective score is compared with both the decision-level fusion admission score threshold and the data-level fusion admission score threshold. If a modal effective score is greater than both the decision-level fusion admission score threshold and the data-level fusion admission score threshold, the sensing sensor corresponding to that modal effective score can function as both a decision-level fusion sensor and a data-level fusion sensor. If a modal effective score is only greater than the decision-level fusion admission score threshold, the sensing sensor corresponding to that modal effective score only functions as a decision-level fusion sensor. If a modal effective score is only greater than the data-level fusion admission score threshold, the sensing sensor corresponding to that modal effective score only functions as a data-level fusion sensor. If the effective score of a certain modality is simultaneously less than both the decision-level fusion admission score threshold and the data-level fusion admission score threshold, it indicates that the sensing sensor corresponding to that modality's effective score is severely affected by environmental interference, resulting in erroneous detection results and excessively noisy raw data. Therefore, it is directly removed from both fusion branches and does not participate in any fusion calculations. For example, if the effective modality score of a visible light camera in a rain or fog scene is 10, then both branches are disabled.

[0043] In step 103, obstacle detection is performed on the perception data of each decision-level fusion sensor, and the detection results are fused to obtain the first fusion detection result.

[0044] In step 104, the sensing data from each data-level fusion sensor are fused to generate fused data, and a second fusion detection result is obtained based on the fused data.

[0045] In this embodiment, after determining the decision-level fusion sensor and the data-level fusion sensor, for the decision-level fusion sensor, the independent AI detection model of each sensor can be invoked to output obstacle detection results and target confidence scores, and weighted fusion is performed on these to obtain the first fusion detection result. For the data-level fusion sensor, the original perception data of each sensor can be read, and after completing spatiotemporal synchronization and coordinate registration, multimodal raw data fusion is performed. The fused unified data is then sent to the multimodal detection network to output the second fusion detection result.

[0046] In step 105, autonomous obstacle avoidance path planning for the target low-altitude aircraft is achieved based on the first fusion detection result and the second fusion detection result.

[0047] In this embodiment, after obtaining the first fusion detection result and the second fusion detection result based on the two branches, the obstacle targets in the first and second fusion detection results can be matched, the obstacle position, size and confidence level can be weighted and corrected, and the final obstacle information can be output and sent to the aircraft obstacle avoidance planning module.

[0048] Optionally, step 105 may include: The obstacles in the first fusion detection result and the obstacles in the second fusion detection result are matched to obtain the target association matching result.

[0049] For target obstacles that are successfully matched in the target association matching results, the first confidence weight corresponding to the first fusion detection result and the second confidence weight of the second fusion detection result are determined based on the current flight scene label, and the information of the target obstacle is fused according to the first confidence weight and the second confidence weight.

[0050] Based on the information of other obstacles besides the target obstacle in the target association matching results and the fused target obstacle information, the autonomous obstacle avoidance path planning of the target low-altitude aircraft is realized.

[0051] Specifically, target association matching addresses the issues of duplicate detection of the same obstacle and target misalignment in the two-path fusion results, distinguishing between three types of targets: common obstacles, obstacles detected only by the decision branch, and small obstacles detected only by the data-level branch. The specific process may include: 1. Unified coordinate transformation: The pixel boxes of all obstacles in the two results are uniformly converted into three-dimensional bounding boxes in the aircraft body coordinate system, eliminating the difference in coordinate dimensions between the image plane and the fused point cloud.

[0052] 2. Multi-dimensional matching criteria: The system simultaneously employs three conditions for matching: Intersection over Union (IOU) of the three-dimensional bounding box, Euclidean distance between obstacle centers, and obstacle category. If the IOU is greater than or equal to a set threshold, the center distance is less than a safety threshold, and the obstacle category is consistent, the obstacle is determined to be the same obstacle (a common target). If the matching conditions are not met, the obstacle is determined to be a unique target on the branch.

[0053] Then, a dual-channel joint correction is performed on the matched common obstacles (i.e., target obstacles). The first confidence weight and the second confidence weight can be obtained through a preset scene weight table. For example, in a scene with dense cables: the first fusion detection result has a weight of 0.3 and the second fusion detection result has a weight of 0.7, focusing on accepting small targets in data-level branches; in a clear sky and long-distance cruise: the first fusion detection result has a weight of 0.7 and the second fusion detection result has a weight of 0.3, focusing on low-latency decision branches; in a rainy or foggy environment: the two weights are balanced at 0.5, and they are mutually verified to suppress false detections.

[0054] Furthermore, for obstacles detected only by the first fusion: considering the lack of underlying fusion information for corroboration, all parameters of the target can be directly retained, with the confidence level appropriately lowered. For small obstacles detected only by the second fusion: considering the in-depth mining of underlying data, the cable-like target can be fully retained, with the base confidence level appropriately increased, and marked as a high-risk small obstacle. Then, global false detection filtering is performed: a minimum comprehensive confidence threshold is set, and obstacles with a corrected confidence level below the threshold are directly removed, eliminating false targets generated by the two fusion methods. Afterward, all corrected valid obstacles can be encapsulated into a standardized global obstacle data package, serving as the sole perception input for the obstacle avoidance planning module.

[0055] This invention first acquires perception data collected by various sensing sensors on a target low-altitude aircraft, and then performs scene recognition based on the perception data to obtain a current flight scene label. Next, based on the current flight scene label, the sensor types of each sensing sensor are classified to determine decision-level fusion sensors and data-level fusion sensors. Then, obstacle detection is performed on the perception data of each decision-level fusion sensor, and the detection results are fused to obtain a first fusion detection result. The perception data of each data-level fusion sensor are then fused to generate fusion data, and a second fusion detection result is obtained based on the fusion data. Finally, autonomous obstacle avoidance path planning for the target low-altitude aircraft is implemented based on the first and second fusion detection results. By identifying sensors with high reliability under the current flight scenario label through scene recognition, decision-level fusion sensors can be selected to achieve rapid perception of macroscopic obstacles. At the same time, sensors adapted to the current environment (i.e., corresponding to the current flight scenario label) can be selected as data-level fusion sensors to fuse raw perception data, thereby fully mining the weak features of small obstacles. The feature representation capability is improved by fusing the detection results of the two paths, reducing the probability of false detection or missed detection, and thus balancing the real-time performance and detection accuracy of obstacle detection. This provides support for achieving high-precision and high-reliability autonomous obstacle avoidance for low-altitude aircraft.

[0056] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0057] The following are system embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0058] Figure 2 The diagram illustrates the structure of an AI-based fusion perception adaptive obstacle avoidance system for low-altitude aircraft provided in an embodiment of the present invention. For ease of explanation, only the parts relevant to the embodiment of the present invention are shown, and are described in detail below: like Figure 2 As shown, the AI-based adaptive obstacle avoidance system for low-altitude aircraft includes: The first processing module 21 is used to acquire the perception data collected by each perception sensor on the target low-altitude aircraft, and to perform scene recognition based on the perception data to obtain the current flight scene label.

[0059] The second processing module 22 is used to classify the sensor types of each perception sensor based on the current flight scenario label, and to determine the decision-level fusion sensor and the data-level fusion sensor.

[0060] The first fusion module 23 is used to perform obstacle detection on the perception data of each decision-level fusion sensor, and fuse the detection results to obtain the first fusion detection result.

[0061] The second fusion module 24 is used to fuse the sensing data of each data-level fusion sensor to generate fused data and obtain the second fusion detection result based on the fused data.

[0062] The adaptive obstacle avoidance module 25 is used to realize the autonomous obstacle avoidance path planning of the target low-altitude aircraft based on the first fusion detection result and the second fusion detection result.

[0063] In one possible implementation, the first processing module 21 is specifically used for: Environmental statistical features of each sensing sensor are extracted based on the sensing data.

[0064] Scene identification is performed based on environmental statistical characteristics to obtain the current flight scene label.

[0065] In one possible implementation, the first processing module 21 is specifically used for: Scene categories are identified based on environmental statistical characteristics to obtain the current scene category label.

[0066] Based on the current major scene labels, the effective sensors among the various perception sensors are selected, and obstacle detection is performed on the perception data of each effective sensor. Based on the detection results, the scene is further subdivided and identified to obtain the current flight scene label.

[0067] In one possible implementation, the first processing module 21 is specifically used for: Scene identification is performed based on environmental statistical characteristics to obtain the first flight scene label.

[0068] Obstacle detection is performed on each sensing data, and the structured features of each detection result are extracted. Scene recognition is then performed based on the structured features to obtain a second flight scene label.

[0069] Obtain the current flight scenario label based on the first flight scenario label and the second flight scenario marker.

[0070] In one possible implementation, the second processing module 22 is specifically used for: Based on the current flight scenario labels and the preset scenario and sensor adaptation rule library, the basic scenario adaptation score of each perception sensor is obtained.

[0071] Environmental statistical features of each sensing sensor are extracted based on the various sensing data, and dynamic correction scores of each sensing sensor are calculated based on these environmental statistical features.

[0072] The corresponding basic scene adaptation score is corrected based on the dynamic correction score of each sensing sensor to obtain the effective modal score of each sensing sensor.

[0073] The effective scores of each modality are compared with the decision-level fusion admission score threshold and the data-level fusion admission score threshold, respectively. Based on the comparison results, the sensor types of each sensing sensor are classified to determine the decision-level fusion sensor and the data-level fusion sensor.

[0074] In one possible implementation, the adaptive obstacle avoidance module 25 is specifically used for: The obstacles in the first fusion detection result and the obstacles in the second fusion detection result are matched to obtain the target association matching result.

[0075] For target obstacles that are successfully matched in the target association matching results, the first confidence weight corresponding to the first fusion detection result and the second confidence weight of the second fusion detection result are determined based on the current flight scene label, and the information of the target obstacle is fused according to the first confidence weight and the second confidence weight.

[0076] Based on the information of other obstacles besides the target obstacle in the target association matching results and the fused target obstacle information, the autonomous obstacle avoidance path planning of the target low-altitude aircraft is realized.

[0077] Figure 3 This is a schematic diagram of a low-altitude aircraft based on AI-fused perception, provided in an embodiment of the present invention. Figure 3 As shown, the AI-based fusion perception low-altitude aircraft 3 of this embodiment includes a processor 30 and a memory 31. The memory 31 stores a computer program 32. When the processor 30 executes the computer program 32, it implements the steps in the various method embodiments described above. Alternatively, when the processor 30 executes the computer program 32, it implements the functions of each module / unit in the system embodiments described above.

[0078] For example, computer program 32 can be divided into one or more modules / units, which are stored in memory 31 and executed by processor 30 to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of computer program 32 in AI-based fusion perception low-altitude aircraft 3.

[0079] The low-altitude aircraft 3 based on AI-integrated perception may include, but is not limited to, a processor 30 and a memory 31. Those skilled in the art will understand that... Figure 3This is merely an example of a low-altitude aircraft 3 based on AI fusion perception, and does not constitute a limitation on the low-altitude aircraft 3 based on AI fusion perception. It may include more or fewer components than shown in the figure, or combine certain components, or different components. For example, the low-altitude aircraft 3 based on AI fusion perception may also include input / output devices, network access devices, buses, etc.

[0080] The processor 30 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0081] The memory 31 can be an internal storage unit of the AI-based fusion perception low-altitude aircraft 3, such as a hard drive or memory. The memory 31 can also be an external storage device of the AI-based fusion perception low-altitude aircraft 3, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the AI-based fusion perception low-altitude aircraft 3. Furthermore, the memory 31 can include both internal storage units and external storage devices of the AI-based fusion perception low-altitude aircraft 3. The memory 31 is used to store the computer program 32 and other programs and data required by the AI-based fusion perception low-altitude aircraft 3. The memory 31 can also be used to temporarily store data that has been output or will be output.

[0082] For the sake of simplicity and clarity, only the above-described functional modules / units are used as examples. In practical applications, the functions described above can be assigned to different functional modules / units as needed. These modules / units can be implemented in hardware, software, or a combination of both.

[0083] This invention also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the methods described in the above-described method embodiments.

[0084] This invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the methods described in the above-described method embodiments.

[0085] Computer programs include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Computer-readable media can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0086] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0087] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An adaptive obstacle avoidance method for low-altitude aircraft based on AI-fusion perception, characterized in that, include: Acquire the perception data collected by each perception sensor on the target low-altitude aircraft, and perform scene recognition based on the perception data to obtain the current flight scene label; Based on the current flight scenario label, the sensor types of each of the aforementioned sensing sensors are classified to determine the decision-level fusion sensor and the data-level fusion sensor. Obstacle detection is performed on the perception data of each of the decision-level fusion sensors, and the detection results are fused to obtain the first fusion detection result; The sensed data from each of the data-level fusion sensors are fused to generate fused data, and a second fusion detection result is obtained based on the fused data; Based on the first fusion detection result and the second fusion detection result, the autonomous obstacle avoidance path planning of the target low-altitude aircraft is realized.

2. The adaptive obstacle avoidance method for low-altitude aircraft based on AI fusion perception according to claim 1, characterized in that, Based on the aforementioned perception data, scene recognition is performed to obtain the current flight scene label, including: Environmental statistical features of each of the aforementioned sensing sensors are extracted based on the sensing data. Scene identification is performed based on the environmental statistical characteristics to obtain the current flight scene label.

3. The adaptive obstacle avoidance method for low-altitude aircraft based on AI fusion perception according to claim 2, characterized in that, Scene recognition is performed based on the aforementioned environmental statistical features to obtain the current flight scene label, including: Based on the environmental statistical features, scene category identification is performed to obtain the current scene category label; Based on the current major scene labels, select the effective sensors among the various perception sensors, and perform obstacle detection on the perception data of each effective sensor. Based on the detection results, perform subdivided scene recognition to obtain the current flight scene label.

4. The adaptive obstacle avoidance method for low-altitude aircraft based on AI fusion perception according to claim 2, characterized in that, Scene recognition is performed based on the aforementioned environmental statistical features to obtain the current flight scene label, including: Based on the environmental statistical features, scene recognition is performed to obtain a first flight scene label; Obstacle detection is performed on each of the aforementioned perception data, and the structured features of each detection result are extracted. Scene recognition is then performed based on the structured features to obtain a second flight scene label. The current flight scenario label is obtained based on the first flight scenario label and the second flight scenario marker.

5. The adaptive obstacle avoidance method for low-altitude aircraft based on AI fusion perception according to claim 1, characterized in that, The step of classifying each of the perception sensors based on the current flight scenario label, and determining the decision-level fusion sensor and the data-level fusion sensor, includes: Based on the current flight scenario label and the preset scenario and sensor adaptation rule base, the basic scenario adaptation score of each of the sensing sensors is obtained; Based on the sensing data, the environmental statistical features of each sensing sensor are extracted, and the dynamic correction score of each sensing sensor is calculated based on the environmental statistical features. The corresponding basic scene adaptation score is corrected based on the dynamic correction score of each of the sensing sensors to obtain the modal effective score of each of the sensing sensors. The effective scores of each modality are compared with the decision-level fusion admission score threshold and the data-level fusion admission score threshold, respectively. Based on the comparison results, the sensor types of each sensing sensor are classified to determine the decision-level fusion sensor and the data-level fusion sensor.

6. The adaptive obstacle avoidance method for low-altitude aircraft based on AI fusion perception according to claim 1, characterized in that, Based on the first fusion detection result and the second fusion detection result, the autonomous obstacle avoidance path planning of the target low-altitude aircraft is realized, including: Target association matching is performed on the obstacles in the first fusion detection result and the obstacles in the second fusion detection result to obtain the target association matching result; For the target obstacles that are successfully matched in the target association matching results, a first confidence weight and a second confidence weight of the second fusion detection results are determined based on the current flight scene label, and the information of the target obstacles is fused according to the first confidence weight and the second confidence weight; Based on the information of other obstacles besides the target obstacle in the target association matching result and the fused target obstacle information, the autonomous obstacle avoidance path planning of the target low-altitude aircraft is realized.

7. An adaptive obstacle avoidance system for low-altitude aircraft based on AI-fusion perception, characterized in that, include: The first processing module is used to acquire the perception data collected by each perception sensor on the target low-altitude aircraft, and to perform scene recognition based on the perception data to obtain the current flight scene label. The second processing module is used to classify the sensor types of each of the perception sensors based on the current flight scenario label, and to determine the decision-level fusion sensor and the data-level fusion sensor. The first fusion module is used to perform obstacle detection on the perception data of each of the decision-level fusion sensors, and fuse the detection results to obtain the first fusion detection result; The second fusion module is used to fuse the sensing data of each of the data-level fusion sensors to generate fused data, and to obtain a second fusion detection result based on the fused data; An adaptive obstacle avoidance module is used to realize autonomous obstacle avoidance path planning for the target low-altitude aircraft based on the first fusion detection result and the second fusion detection result.

8. A low-altitude aircraft based on AI-integrated perception, characterized in that, It includes a memory and a processor, the memory being used to store a computer program, and the processor being used to call and run the computer program stored in the memory to perform the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7 above.

10. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 6.

Images