Obstacle recognition method and aircraft

Abstract

The application relates to the technical field of aircrafts, and provides an obstacle identification method and an aircraft. Wherein, a voxel occupancy distribution of a flight space where the aircraft is located is predicted by using a current circumferential image of the aircraft, and a candidate reflection region is generated based on the voxel occupancy distribution, so as to focus on a candidate obstacle in the candidate reflection region to perform arbitration of mirror image misjudgment. Wherein, the arbitration process first eliminates the candidate obstacle without occlusion between the aircraft, and takes the remaining candidate obstacle as a mirror image misjudgment candidate obstacle; then, a one-to-one binding relationship between the mirror image misjudgment candidate obstacle and a real obstacle is established by using multi-dimensional state information; finally, the real identity of the mirror image misjudgment candidate obstacle is confirmed through consistency checking, so that accurate judgment of the mirror image misjudgment is realized, the obstacle identification precision is improved, and the flight stability and safety of the aircraft are effectively guaranteed.

Description

Technical Field

[0001] This application relates to the field of aircraft technology, and in particular to an obstacle recognition method and an aircraft. Background Technology

[0002] An aircraft is an aerial vehicle capable of autonomous or remote-controlled flight, widely used in aerial photography, surveying, logistics, and inspection. During flight, the aircraft must continuously perceive its surroundings to identify obstacles, ensuring flight safety and successful mission execution. In low-altitude urban flight scenarios, aircraft typically fly at altitudes below 150 meters, and are close to tall buildings. Many office buildings, commercial complexes, and high-rise residential buildings utilize glass curtain walls or high-reflectivity facades. Please refer to [reference needed]. Figure 1 When an aircraft approaches such a building, it may observe the specular reflection of itself, other aircraft, ground vehicles, or other real obstacles, causing the aircraft to identify such specular reflections as obstacles, thus affecting the accuracy of obstacle identification. Summary of the Invention

[0003] This application provides an obstacle recognition method and an aircraft, which can accurately determine false detections of mirror images, improve obstacle recognition accuracy, and effectively ensure the flight stability and safety of the aircraft.

[0004] Firstly, the obstacle recognition method provided in this application includes: Acquire the current circumferential image of the aircraft, perform 3D occupancy prediction on the current circumferential image, and obtain the voxel occupancy distribution of the flight space in which the aircraft is located; Voxels on the building surface are extracted from the voxel occupancy distribution, and candidate reflection regions are generated based on the voxels on the building surface. Candidate obstacles in the flight space are identified based on the voxel occupancy distribution, and candidate obstacles located outside the candidate reflection area are marked as real obstacles; For each candidate obstacle located within the candidate reflection area, the occlusion voxel between the aircraft and the candidate obstacle is extracted from the voxel occupancy distribution, and the occlusion voxel is used to identify whether the candidate obstacle is occluded. If the candidate obstacle is identified as occluded, it is marked as a mirror false detection candidate obstacle; otherwise, it is marked as a real obstacle. Acquire multi-dimensional state information of the aircraft and multi-dimensional state information of real obstacles; For each mirrored false detection candidate obstacle, obtain the multidimensional state information of the mirrored false detection candidate obstacle, and calculate the multidimensional state information of the assumed real obstacle corresponding to the mirrored false detection candidate obstacle based on the candidate reflection area corresponding to the mirrored false detection candidate obstacle; establish a one-to-one binding relationship between the mirrored false detection candidate obstacle and the aircraft or real obstacle based on the multidimensional state information of the assumed real obstacle, the multidimensional state information of the aircraft, and the multidimensional state information of the real obstacle; For each mirrored false detection candidate obstacle, verify whether the one-to-one binding relationship corresponding to the mirrored false detection candidate obstacle is continuously established within the time window of the current circumferential image; if it is continuously established, then the mirrored false detection candidate obstacle is determined to be a mirrored false detection obstacle.

[0005] Secondly, the aircraft provided in this application includes: ontology; The drive mechanism is used to propel the aircraft into flight. The memory, located within the main unit, is used to store computer programs; The processor, located in the main body, is used to execute computer programs to implement the obstacle recognition method provided in this application.

[0006] The obstacle recognition scheme provided in this application predicts the voxel occupancy distribution of the flight space where the aircraft is located using the current circumferential image of the aircraft, and generates candidate reflection regions based on the voxel occupancy distribution. This allows for the arbitration of mirror-image false detections by focusing on candidate obstacles within these reflection regions. The arbitration process first eliminates candidate obstacles that are not obstructed by the aircraft, treating the remaining candidate obstacles as mirror-image false detection candidates. Then, using multi-dimensional state information, a one-to-one binding relationship is established between the mirror-image false detection candidate obstacles and the real obstacles. Finally, a consistency check confirms the true identity of the mirror-image false detection candidate obstacles, thereby achieving accurate judgment of mirror-image false detections, improving obstacle recognition accuracy, and effectively ensuring the flight stability and safety of the aircraft. Attached Figure Description

[0007] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is an example diagram showing the specular reflection of a virtual image created by an aircraft or obstacle on a building. Figure 2 This is a flowchart illustrating the obstacle recognition method provided in an embodiment of this application; Figure 3 This is a detailed flowchart illustrating the process of generating candidate reflection regions in an embodiment of this application; Figure 4 This is a detailed flowchart illustrating the process of identifying whether a candidate obstacle is occluded in an embodiment of this application. Figure 5 yes Figure 4 Detailed process diagram of S1420; Figure 6 This is a detailed flowchart illustrating obstacle matching in an embodiment of this application; Figure 7 This is a schematic diagram of the obstacle recognition device provided in the embodiments of this application; Figure 8 This is a schematic diagram of the structure of the aircraft provided in the embodiments of this application; Figure 9 This is an example diagram of the product form of the aircraft provided in the embodiments of this application. Detailed Implementation

[0009] To make the technical problems, technical solutions, and beneficial effects solved by this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0010] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0011] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0012] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0013] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0014] References to "one embodiment" or "some embodiments" in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0015] This application provides an obstacle recognition method and an aircraft. The obstacle recognition method includes: acquiring a current circumferential image of the aircraft; performing three-dimensional occupancy prediction on the current circumferential image to obtain a voxel occupancy distribution of the flight space where the aircraft is located; extracting building surface voxels from the voxel occupancy distribution and generating candidate reflection regions based on the building surface voxels; identifying candidate obstacles in the flight space based on the voxel occupancy distribution, and marking candidate obstacles located outside the candidate reflection regions as real obstacles; for each candidate obstacle located within a candidate reflection region, extracting occlusion voxels located between the aircraft and the candidate obstacle from the voxel occupancy distribution, and identifying whether the candidate obstacle is occluded based on the occlusion voxels; if the candidate obstacle is identified as occluded, marking the candidate obstacle as a mirror false detection candidate obstacle, otherwise marking it as a real obstacle. The system acquires multidimensional state information of the aircraft and real obstacles. For each mirrored false alarm candidate obstacle, it acquires the multidimensional state information of the mirrored false alarm candidate obstacle and calculates the multidimensional state information of the assumed real obstacle corresponding to the mirrored false alarm candidate obstacle based on the candidate reflection area corresponding to the mirrored false alarm candidate obstacle. Based on the multidimensional state information of the assumed real obstacle, the multidimensional state information of the aircraft, and the multidimensional state information of the real obstacle, it establishes a one-to-one binding relationship between the mirrored false alarm candidate obstacle and the aircraft or real obstacle. For each mirrored false alarm candidate obstacle, it verifies whether the one-to-one binding relationship corresponding to the mirrored false alarm candidate obstacle is continuously valid within the time window of the current circumferential image. If it is continuously valid, the mirrored false alarm candidate obstacle is determined to be a mirrored false alarm obstacle.

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] Please refer to Figure 2 , Figure 2 This is a flowchart illustrating an obstacle recognition method provided in an embodiment of this application, as shown below. Figure 2 As shown, the process of this obstacle recognition method can be as follows: In S110, the current circumferential image of the aircraft is acquired, and three-dimensional occupancy prediction is performed on the current circumferential image to obtain the voxel occupancy distribution of the flight space in which the aircraft is located.

[0018] An aircraft is an aerial vehicle capable of autonomous flight or remote control. It can perform flight maneuvers such as takeoff, hovering, forward movement, and landing through propellers, jets, or other types of drive mechanisms, and is typically powered by fuel or electricity. For example, in the embodiments of this application, the aircraft can be an electric vertical takeoff and landing (eVTOL) aircraft focused on low-altitude flight, with typical applications including urban air traffic, logistics delivery, and emergency rescue.

[0019] In this embodiment, an occupancy prediction model is pre-trained. This model is configured to take a two-dimensional circumferential image as input and a three-dimensional voxel occupancy distribution as output. The voxel occupancy distribution represents the occupancy probability and semantic category of each voxel in the source space of the circumferential image in the form of a three-dimensional grid. Each voxel is a three-dimensional cube with a side length of a preset resolution. The occupancy probability reflects the likelihood of an obstacle existing at the corresponding spatial location of the voxel, and the semantic category identifies the category of obstacle to which the voxel belongs, such as buildings, trees, or other aircraft. It should be noted that this embodiment does not limit the specific network structure of the occupancy prediction model. It can be built based on a three-dimensional convolutional neural network, graph convolutional neural network, or Transformer architecture, as long as it can effectively map the circumferential image to the three-dimensional voxel occupancy distribution. Furthermore, this embodiment does not limit the model training method and can adopt supervised learning, self-supervised learning, or semi-supervised learning strategies.

[0020] For example, firstly, an image sample set containing a large number of real flight environment circumferential image samples is acquired. For each circumferential image sample in the image sample set, the ground truth value of the labeled voxel occupancy distribution (including the occupancy state, semantic category label, and multi-dimensional state information label of each voxel, where the multi-dimensional state information label includes, but is not limited to, dynamic attributes such as velocity and orientation) is obtained as the training target for supervised learning. Then, the circumferential image samples are input into the initial occupancy prediction model to obtain the predicted voxel occupancy distribution. By calculating the multi-task loss between the predicted voxel occupancy distribution and the ground truth value (covering occupancy probability cross-entropy, semantic category cross-entropy, and spatial state regression loss), the model parameters are optimized through backpropagation, and the training is iteratively performed until the training convergence condition is met. After training, the occupancy prediction model can receive circumferential images acquired by the surround-view camera on the aircraft and output the corresponding three-dimensional space voxel occupancy distribution, covering information such as occupancy probability, semantic category, and multi-dimensional spatial state.

[0021] The obstacle recognition method provided in this application will be described in detail below, based on the occupancy prediction model provided in this application.

[0022] During actual flight, the aircraft acquires circumferential images of its current flight space using its onboard surround-view cameras, which are recorded as the current circumferential image. For example, at time t, the circumferential image of the flight space acquired by the aircraft using its onboard surround-view cameras (e.g., composed of multiple wide-angle cameras arranged circumferentially, simultaneously capturing 360° panoramic images and stitching them together into a circumferential image in a unified coordinate system) is the current circumferential image at time t, while the circumferential images acquired at times prior to time t (e.g., time t-1) are the historical circumferential images at time t.

[0023] As shown above, after acquiring the current circumferential image, the aircraft inputs the current circumferential image into the occupancy prediction model to perform three-dimensional occupancy prediction, thereby obtaining the voxel occupancy distribution of the flight space in which the aircraft is located.

[0024] In S120, voxels on the building surface are extracted from the voxel occupancy distribution, and candidate reflection regions are generated based on the voxels on the building surface.

[0025] After acquiring the voxel occupancy distribution of its flight space, the aircraft first filters out building voxels belonging to the "building" category based on semantic classification. Then, it extracts building surface voxels from the building voxels, which are the building voxels located on the outermost layer of the building entity and adjacent to non-occupying voxels. These surface voxels constitute potential specular reflection areas. Subsequently, based on the normal vectors and position information of the building surface voxels, the aircraft generates candidate reflection areas. Each candidate reflection area corresponds to a local planar region composed of continuous building surface voxels.

[0026] In other embodiments, if the semantic category further includes subdivided building subcategories (such as “glass curtain wall”, “metal roof”, “concrete wall”, etc.), then when the aircraft extracts building surface voxels from building voxels, it can extract only building surface voxels with high reflectivity, such as “glass curtain wall” and “metal roof”, to generate candidate reflection areas, thereby improving the physical rationality of the candidate reflection areas and the accuracy of subsequent obstacle recognition.

[0027] In S130, candidate obstacles in the flight space are identified based on the voxel occupancy distribution, and candidate obstacles located outside the candidate reflection area are marked as real obstacles.

[0028] After generating candidate reflection regions using voxel occupancy distribution, the aircraft further utilizes the voxel occupancy distribution for initial obstacle identification. First, the aircraft identifies all candidate obstacles based on voxels with an occupancy probability higher than a probability threshold and a semantic category of "obstacle." A candidate obstacle can correspond to a 3D cluster composed of one or more consecutively occupying voxels. Then, the aircraft traverses all candidate obstacles, determining whether their spatial position falls within any candidate reflection region. If it does, it is considered a virtual image caused by specular reflection and included in subsequent arbitration; otherwise, it is directly marked as a real obstacle. The probability threshold value is not limited here and can be dynamically set based on actual scene lighting conditions and sensor accuracy.

[0029] The above method of filtering out candidate obstacles that require arbitration by selecting candidate reflection regions can avoid redundant calculations on all candidate obstacles, save computing resources, and improve the overall efficiency of obstacle recognition.

[0030] In S140, for each candidate obstacle located within the candidate reflection area, the occlusion voxel between the aircraft and the candidate obstacle is extracted from the voxel occupancy distribution, and the candidate obstacle is identified as occluded based on the occlusion voxel; if the candidate obstacle is identified as occluded, the candidate obstacle is marked as a mirror false detection candidate obstacle, otherwise it is marked as a real obstacle.

[0031] For each candidate obstacle requiring arbitration (a candidate obstacle located within the candidate reflection area), the aircraft further extracts voxels located between the aircraft and the candidate obstacle and whose occupancy probability is higher than the probability threshold from the voxel occupancy distribution, as the occupancy voxels between the two.

[0032] Subsequently, the aircraft identifies whether the candidate obstacle is obstructed based on the spatial relationship between the obstructing voxel and the candidate obstacle; that is, whether the line-of-sight path from the candidate obstacle to the aircraft is blocked by the obstructing voxel. If the candidate obstacle is identified as obstructed, it is determined that the candidate obstacle may be a virtual image generated by specular reflection and is marked as a false image candidate obstacle for subsequent arbitration; if the candidate obstacle is identified as not obstructed, it is determined to be a real obstacle and is marked as a real obstacle accordingly.

[0033] In S150, multi-dimensional state information of the aircraft and multi-dimensional state information of real obstacles are acquired.

[0034] As described above, after marking the mirrored false detection candidate obstacles, the aircraft further acquires its own multidimensional state information, as well as the multidimensional state information of the previously marked real obstacles. It should be noted that, with the constraint that the dimensions of the multidimensional state information of the two are the same, the specific composition of the multidimensional state information is not limited here. For example, it may include spatial state information such as category, position, speed, and orientation, and / or shape state information such as length, width, height, color, texture, and contour features.

[0035] For example, in this embodiment of the application, the voxel occupancy distribution also includes the multidimensional state information of the voxels. Accordingly, when obtaining the multidimensional state information of a real obstacle, the aircraft can aggregate the multidimensional state information of the real obstacle as a whole based on the multidimensional state information of the voxel clusters corresponding to the real obstacle. The aggregation method is not limited here; it can be weighted average, principal component extraction, or dynamic weighted fusion based on voxel confidence, etc.

[0036] When acquiring the multidimensional status information of an aircraft, it can directly obtain it from its own sensors or navigation system. For example, when the multidimensional status information is configured to include position, speed and orientation, the aircraft can directly obtain position, speed and orientation data from the navigation system to complete the acquisition of its own multidimensional status information.

[0037] In S160, for each mirrored false detection candidate obstacle, the multidimensional state information of the mirrored false detection candidate obstacle is obtained, and the multidimensional state information of the assumed real obstacle corresponding to the mirrored false detection candidate obstacle is calculated based on the candidate reflection area corresponding to the mirrored false detection candidate obstacle; based on the multidimensional state information of the assumed real obstacle, the multidimensional state information of the aircraft, and the multidimensional state information of the real obstacle, a one-to-one binding relationship is established between the mirrored false detection candidate obstacle and the aircraft or the real obstacle.

[0038] For each mirrored false alarm candidate obstacle, the aircraft further acquires its multidimensional state information. Specifically, the aircraft extracts the multidimensional state information of the voxel cluster corresponding to the mirrored false alarm candidate obstacle from the voxel occupancy distribution, and then obtains the overall multidimensional state information of the mirrored false alarm candidate obstacle through aggregation. The aggregation method is not limited here; weighted average, principal component extraction, or dynamic weighted fusion based on voxel confidence can be used. In practice, when acquiring the multidimensional state information of the mirrored false alarm candidate obstacle, the same aggregation method used when acquiring the multidimensional state information of the real obstacle can be adopted to ensure the consistency and comparability of the state information representation.

[0039] Subsequently, based on the multidimensional state information of the falsely detected candidate obstacle in the mirror image, and combined with its corresponding candidate reflection area, the aircraft performs a geometric transformation according to the principle of mirror reflection to calculate the multidimensional state information of the assumed real obstacle corresponding to the falsely detected candidate obstacle in the flight space.

[0040] Finally, the aircraft matches the multidimensional state information of the assumed real obstacle with its own multidimensional state information and the multidimensional state information of known real obstacles to establish a one-to-one binding relationship between the mirrored false obstacle candidate and the aircraft or the real obstacle. For example, if a one-to-one binding relationship is established between a mirrored false obstacle candidate and the aircraft, it indicates that the mirrored false obstacle candidate may be a mirror reflection of the aircraft itself.

[0041] It should be noted that in real-world scenarios, a physical object may generate multiple mirror images on multiple reflective surfaces. Therefore, when performing obstacle matching, the aircraft uses each generated candidate reflection area as an anchor point to perform independent binding judgment on the mirror image misdetected candidate obstacle corresponding to each candidate reflection area, so as to avoid confusion across reflection surfaces.

[0042] It should be noted that in other embodiments, in addition to the above-mentioned candidate reflection area and occlusion determination, real obstacles can also be cross-confirmed by combining data from other sensors such as radar sensors. For example, if the same obstacle is detected simultaneously by lidar and millimeter-wave radar, then the obstacle can be marked as a real obstacle.

[0043] In S170, for each mirror false detection candidate obstacle, it is verified whether the one-to-one binding relationship corresponding to the mirror false detection candidate obstacle is continuously established within the time window of the current circumferential image; if it is continuously established, the mirror false detection candidate obstacle is determined to be a mirror false detection obstacle.

[0044] The current circumferential image's time window refers to the continuous frame interval formed with the current frame as the reference. For example, it can be the continuous frame interval formed by extending N frames backward from the current frame, or the continuous frame interval formed by tracing back N frames from the current frame as the end frame, or the continuous frame interval formed by extending M frames before and after the current frame as the center, and so on.

[0045] During the specific verification process, the aircraft continuously tracks the stability of the binding relationship within the time window. For each falsely detected obstacle, the aircraft can calculate the percentage of frames in which the one-to-one binding relationship between the assumed real obstacle and the aircraft or a known real obstacle is established within the time window. If this percentage is greater than or equal to a threshold, the binding relationship is considered to be continuously established within the time window. For example, assuming a time window length of 10 frames and a threshold of 80%, if the one-to-one binding relationship between the assumed real obstacle and the aircraft or a known real obstacle is established in more than 8 of those frames, the binding relationship is considered to be continuously established. The value of the threshold is not limited here and can be dynamically adjusted based on factors such as the aircraft's speed, image acquisition frame rate, and environmental complexity. For example, in high-speed flight or strong reflection interference scenarios, the threshold can be appropriately increased to 90% to enhance the robustness of the criterion; while in low-speed or sparsely reflective scenarios, the threshold can be lowered to 70% to improve detection sensitivity, and so on.

[0046] If the one-to-one binding relationship between the assumed real obstacle and the aircraft or known real obstacle corresponding to a mirrored false obstacle candidate remains valid within the time window, the aircraft determines that the mirrored false obstacle candidate is a virtual image generated by the bound aircraft or real obstacle in the mirror reflection, and marks the mirrored false obstacle candidate as a mirrored false obstacle. If the one-to-one binding relationship between the assumed real obstacle and the aircraft or known real obstacle corresponding to a mirrored false obstacle candidate does not remain valid within the time window, the mirrored false obstacle candidate is marked as a suspected obstacle, and a secondary verification process is triggered to further verify whether the one-to-one binding remains valid in subsequent time windows after the initial time window. If yes, it is updated to a mirrored false obstacle; otherwise, it is updated to a real obstacle.

[0047] As described above, the obstacle recognition scheme provided in this application predicts the voxel occupancy distribution of the flight space where the aircraft is located using the current circumferential image of the aircraft, and generates candidate reflection regions based on this voxel occupancy distribution. This allows for the arbitration of mirror-image false detections by focusing on candidate obstacles within these reflection regions. Specifically, this arbitration process first eliminates candidate obstacles that are not obstructed by the aircraft, treating the remaining candidate obstacles as mirror-image false detection candidates. Then, using multi-dimensional state information, a one-to-one binding relationship is established between the mirror-image false detection candidate obstacles and the real obstacles. Finally, a consistency check confirms the true identity of the mirror-image false detection candidate obstacles, thereby achieving accurate judgment of mirror-image false detections, improving obstacle recognition accuracy, and effectively ensuring the flight stability and safety of the aircraft.

[0048] Alternatively, in one embodiment, please refer to Figure 3 Candidate reflection regions are generated based on voxels on the building surface, including: In S1210, the building surface contour is extracted based on the building surface voxels, and a plane is fitted based on the center point of the building surface voxels to obtain the fitted plane. In S1220, candidate reflection regions are generated based on the building surface profile and the fitted plane.

[0049] The aircraft first extracts the position (i.e., the three-dimensional coordinates of the center point) and normal vector of each building surface voxel. Then, based on the normal vector and position of each building surface voxel, it performs voxel clustering, grouping building surface voxels with similar normal vectors and spatial proximity into the same building surface unit.

[0050] For each building surface unit, the aircraft fits the boundary contour of the building surface unit based on the building surface voxels within that unit to obtain the building surface contour. Simultaneously, using the center points of all building surface voxels within the unit, a plane fitting is performed to obtain the fitting plane for that building surface unit. Subsequently, based on the building surface contour and the fitting plane, candidate reflection regions with consistent geometric characteristics are generated. The boundary of these candidate reflection regions is obtained by projecting the building surface contour onto the fitting plane, and its normal vector is consistent with the unit normal vector of the fitting plane. This ensures that the candidate reflection regions possess physical interpretability and geometric fidelity in three-dimensional space, thus providing a high-confidence geometric constraint basis for subsequent false image detection arbitration. The method of plane fitting is not limited here; for example, geometric construction methods such as least squares, RANSAC, or principal component analysis can be used for plane fitting.

[0051] For a given building surface element, the fitted plane corresponding to that building surface element can be characterized by a plane equation: n·x+d= 0; Where n is the unit normal vector of the fitted plane, representing the overall orientation of the building surface unit; d is the directed distance from the plane to the origin of the coordinate system, i.e., the plane offset; the sign and magnitude of the value obtained by substituting any point x into the plane equation can determine its spatial position and distance relative to the fitted plane. When the value obtained by substituting is 0, it means that the point is located on the fitted plane; when the value obtained by substituting is positive, it means that the point is located on the side of the positive direction of the normal vector of the fitted plane; when the value obtained by substituting is negative, it means that the point is located on the side of the opposite direction of the normal vector of the fitted plane.

[0052] Optionally, in one embodiment, before generating candidate reflection regions based on the building surface profile and the fitted plane, the method further includes: Obtain the fitting residuals of the fitted plane; Based on the building surface profile and fitted plane, candidate reflection regions are generated, including: If the fitting residual is less than the residual threshold, then candidate reflection areas are generated based on the building surface profile and the fitting plane.

[0053] In this embodiment of the application, after the aircraft obtains the building surface profile and fitting plane of a building surface unit by fitting, it does not immediately generate the candidate reflection area of ​​the building surface unit. Instead, it first evaluates the fitting quality and only generates the candidate reflection area when the fitting quality meets the standard.

[0054] For a given building surface unit, after fitting the building surface contour and the fitted plane, the aircraft further calculates the fitting residual of the fitted plane. The calculation method for the fitting residual is not limited; for example, geometric values ​​such as the mean, median, or maximum Euclidean distance from the center point of all building surface voxels within the building surface unit to the fitted plane can be used as the fitting residual. Subsequently, the aircraft compares the fitting residual with a preset residual threshold. If the fitting residual is less than or equal to the residual threshold, it indicates that the building surface unit is approximately planar, with a stable and reliable geometric structure. In this case, candidate reflection regions are generated based on the building surface contour and the fitted plane of the building surface unit. If the fitting residual is greater than the residual threshold, it indicates that the building surface unit has significant curvature and does not meet the planar assumption. In this case, the building surface unit is skipped to avoid misjudgment of reflection regions due to surface distortion, thereby ensuring the geometric constraint accuracy and robustness of subsequent mirror misdetection arbitration.

[0055] Alternatively, in one embodiment, please refer to Figure 4 Extracting occlusion voxels located between the aircraft and the candidate obstacle from the voxel occupancy distribution, and identifying whether the candidate obstacle is occluded based on the occlusion voxels, including: In S1410, the three-dimensional bounding box of the candidate obstacle is obtained, and the observation frustum of the aircraft to the candidate obstacle is constructed based on the three-dimensional bounding box; In S1420, occlusion voxels located within the observation cone are extracted from the voxel occupancy distribution, and the occlusion voxels are used to identify whether the candidate obstacle is occluded.

[0056] For each candidate obstacle located within the candidate reflection region, when identifying whether the candidate obstacle is occluded, the aircraft first obtains the three-dimensional bounding box of the candidate obstacle. For example, it can perform minimum bounding cube fitting based on the voxel cluster corresponding to the candidate obstacle and use the fitted cube as the three-dimensional bounding box of the candidate obstacle; or it can perform principal component analysis on the voxel cluster to determine its length, width, and height principal axes and then construct an axis-aligned three-dimensional bounding box, and so on.

[0057] Subsequently, the aircraft determines the spatial extent of the candidate obstacle based on the three-dimensional bounding box, and constructs an observation cone triggered by the aircraft that covers the three-dimensional bounding box. This observation cone is used to characterize the overall field of view of the aircraft over the candidate obstacle.

[0058] Based on the constructed observation cone, the aircraft further extracts voxels with occupancy probabilities higher than a probability threshold and located within the observation cone from the voxel occupancy distribution. These voxels serve as occlusion voxels between the aircraft and the candidate obstacle. Finally, the aircraft identifies whether the candidate obstacle is occluded based on the extracted occlusion voxels. For example, the aircraft can determine whether the candidate obstacle is occluded based on the spatial distribution characteristics of the occlusion voxels, such as their spatial distribution density and continuity within the observation cone.

[0059] The above method constructs an observation cone from the aircraft to the candidate obstacle and uses only the occlusion voxels within the observation cone for occlusion determination, which can significantly reduce computational overhead and avoid the interference of irrelevant voxels, ensuring the geometric consistency and physical rationality of the occlusion determination results.

[0060] Alternatively, in one embodiment, please refer to Figure 5 The method identifies whether a candidate obstacle is occluded based on the occlusion voxel, including: In S14210, multiple observation line segments from the aircraft to the candidate obstacle are sampled within the observation cone; In S14220, the proportion of line segments that pass through the occlusion voxel among all observed line segments is determined; if the proportion of line segments is greater than or equal to the proportion threshold, the candidate obstacle is determined to be occluded, otherwise the candidate obstacle is determined to be unoccluded.

[0061] For a candidate obstacle, when the aircraft identifies whether the candidate obstacle is occluded based on the occlusion voxels, it first uniformly samples multiple observation line segments originating from the aircraft and pointing towards the surface of the 3D bounding box of the candidate obstacle within the observation cone of the candidate obstacle, according to the configured angular resolution. The value of the angular resolution is not limited here and can be flexibly set according to factors such as the resolution of the voxel occupancy distribution, computational resource constraints, and real-time task requirements. For example, it can be set to 0.5° to 5° to balance accuracy and efficiency.

[0062] The aircraft checks each observation line segment one by one to see if it intersects with the obstructing voxel within the observation cone. If they intersect, they are marked as obstructed line segments. The aircraft also counts the percentage of obstructed line segments out of the total number of observation line segments to quantify the degree of obstruction.

[0063] If the percentage of line segments exceeds a preset threshold, it indicates that the candidate obstacle is significantly obscured and is unlikely to be directly observed by the aircraft. This suggests that the candidate obstacle is more likely to be a virtual image created by mirror reflection than a real obstacle.

[0064] If the percentage of line segments is less than the threshold, it indicates that the candidate obstacle is less obscured and is more likely to be directly observed by the aircraft. This suggests that the candidate obstacle is more likely to be a real obstacle than a virtual image created by a mirror reflection.

[0065] It should be noted that the embodiments of this application do not limit the value of the proportional threshold, and can be flexibly set according to actual task requirements, sensor accuracy and environmental complexity.

[0066] By sampling several observation line segments within the observation cone, the above method transforms occlusion determination into a quantitative assessment of line-of-sight penetration. This not only further reduces computational overhead but also makes the determination results more physically interpretable.

[0067] Alternatively, in one embodiment, please refer to Figure 6 Based on the multidimensional state information of the assumed real obstacle, the multidimensional state information of the aircraft, and the multidimensional state information of the real obstacle, a one-to-one binding relationship is established between the mirrored false detection candidate obstacle and the aircraft or real obstacle, including: In S1610, based on the multidimensional state information of the assumed real obstacle and the multidimensional state information of the aircraft, the category difference, position difference, speed difference and orientation difference between the assumed real obstacle and the aircraft are obtained, and the category difference, position difference, speed difference and orientation difference between the assumed real obstacle and the aircraft are fused to obtain the comprehensive difference between the assumed real obstacle and the aircraft. In S1620, based on the multidimensional state information of the assumed real obstacle and the multidimensional state information of the real obstacle, the category difference, position difference, speed difference and orientation difference between the assumed real obstacle and the real obstacle are obtained, and the category difference, position difference, speed difference and orientation difference between the assumed real obstacle and the real obstacle are fused to obtain the comprehensive difference between the assumed real obstacle and the real obstacle. In S1630, based on the combined differences between the assumed real obstacle and the aircraft and the real obstacle, a one-to-one binding relationship is established between the mirrored false detection candidate obstacle and the aircraft or the real obstacle.

[0068] In this embodiment, a more stable spatial state dimension is used for obstacle matching, avoiding mismatches caused by the shape state dimension being susceptible to interference from lighting, angle and deformation, thereby improving the accuracy of obstacle matching.

[0069] The multidimensional state information acquired by the aircraft is spatial state-dimensional multidimensional state information, including category, position, velocity, and orientation. During matching, for a hypothetical real obstacle, the aircraft obtains the category, position, velocity, and orientation differences between the hypothetical real obstacle and the aircraft based on the multidimensional state information of the hypothetical real obstacle and the aircraft. On this basis, the aircraft normalizes these differences to the same dimension, and then fuses them according to a configured fusion strategy to obtain the comprehensive difference between the hypothetical real obstacle and the real obstacle. The fusion strategy for this comprehensive difference is not limited here; for example, it can take the average, median, or maximum value of the differences in each dimension, or other geometric values.

[0070] Furthermore, based on the multidimensional state information of the hypothetical real obstacle and the real obstacle, the aircraft obtains the differences in category, position, speed, and orientation between the hypothetical real obstacle and the real obstacle. On this basis, the aircraft normalizes these differences to the same dimension, and then fuses them according to a configured fusion strategy to obtain the comprehensive difference between the hypothetical real obstacle and the real obstacle. Similarly, while the fusion strategy for the comprehensive difference between the hypothetical real obstacle and the aircraft is the same, no specific restrictions are placed here on the fusion strategy for the comprehensive difference between the hypothetical real obstacle and the real obstacle.

[0071] Understandably, for a hypothetical real obstacle, the smaller the overall difference between it and the aircraft or real obstacle, the higher the probability that the corresponding false alarm candidate obstacle is a specular reflection of the aircraft or real obstacle. Correspondingly, after acquiring the overall differences between each hypothetical real obstacle and the aircraft, and the overall differences between each hypothetical real obstacle and each real obstacle, the aircraft further assigns an optimal matching target to each hypothetical real obstacle, establishing a one-to-one binding relationship between each hypothetical real obstacle and the aircraft or real obstacle. The specific matching method for this one-to-one binding relationship is not limited here, and may include, but is not limited to, the Hungarian algorithm, greedy matching, or threshold-based minimum difference matching, etc. For example, when using threshold-based minimum difference matching, the aircraft sets a matching threshold. For each assumed real obstacle, it determines the smallest comprehensive difference between itself and the aircraft and each real obstacle. If the minimum comprehensive difference is less than or equal to the matching threshold, the mirror image of the assumed real obstacle is bound to the target object (which can be the aircraft or a real obstacle) corresponding to the minimum comprehensive difference. If the minimum comprehensive difference is greater than the matching threshold, the matching fails.

[0072] Optionally, in one embodiment, the category differences, position differences, speed differences, and orientation differences between the assumed real obstacle and the aircraft are integrated to obtain a comprehensive difference between the assumed real obstacle and the aircraft, including: The weighted summation of the category differences, position differences, speed differences, and orientation differences between the assumed real obstacle and the aircraft yields the comprehensive difference between the assumed real obstacle and the aircraft; where the weight coefficient of each difference item is determined according to the importance of its data dimension in obstacle avoidance decision-making.

[0073] This application provides an optional fusion strategy. The aircraft assigns corresponding weight coefficients to category differences, position differences, speed differences, and orientation differences based on the importance of each difference item's data dimension in obstacle avoidance decision-making. The combined difference between the assumed real obstacle and the aircraft is then obtained by multiplying each difference item by its corresponding weight coefficient and summing the results. This can be expressed as: ; in, This indicates the overall difference between the assumed real obstacle and the aircraft. This indicates the category difference between the assumed real obstacle and the aircraft. This indicates the positional difference between the assumed real obstacle and the aircraft. This indicates the speed difference between the hypothetical real obstacle and the aircraft. This indicates the difference in orientation between the assumed real obstacle and the aircraft. - These are the weight coefficients for the corresponding difference terms, and they satisfy... =1.

[0074] Similarly, in the embodiments of this application, the same weighted summation strategy is used when fusing the comprehensive differences between the assumed real obstacle and the real obstacle, which will not be repeated here.

[0075] Understandably, in obstacle avoidance decision-making, the position, speed, orientation, and type of obstacles have varying degrees of importance. For example, the position of an obstacle directly determines the aircraft's avoidance path and timing, and is usually the most important. The speed and orientation of an obstacle jointly determine its motion trend and collision risk, and are of secondary importance. The type of obstacle has more influence on the aircraft's response strategy selection; for example, birds require immediate avoidance while drones can be anticipated and coordinated, and are of relatively lower importance. Therefore, assigning weight coefficients based on importance allows the fused comprehensive differences to closely match actual obstacle avoidance needs, significantly improving the accuracy of image false detection.

[0076] Optionally, in one embodiment, the obstacle recognition method provided in this application further includes: If, based on the combined differences between the assumed real obstacle and the aircraft and the real obstacle, it fails to establish a one-to-one binding relationship between the mirrored false detection candidate obstacle and the aircraft or the real obstacle, then the mirrored false detection candidate obstacle is marked as a real obstacle.

[0077] Understandably, for a mirrored false alarm candidate obstacle, if its combined difference with the aircraft and all known real obstacles is greater than the matching threshold, it means that the mirrored false alarm candidate obstacle lacks significant correlation with the aircraft or real obstacles in terms of multidimensional features. It is impossible to establish a one-to-one binding relationship between the mirrored false alarm candidate obstacle and the aircraft or real obstacles, indicating that the mirrored false alarm candidate obstacle is more likely to be a real obstacle. Therefore, the mirrored false alarm candidate obstacle is marked as a real obstacle.

[0078] It is understood that, in one embodiment, after determining the mirrored false detection candidate obstacle as a mirrored false detection obstacle, the method further includes: Reduce the obstacle avoidance priority of falsely detected obstacles in obstacle avoidance decisions.

[0079] In this embodiment of the application, for the marked mirror-image falsely detected obstacles, the aircraft further reduces its obstacle avoidance priority in obstacle avoidance decision-making, so that the aircraft prioritizes avoiding real obstacles when making obstacle avoidance decisions, and only initiates emergency avoidance of mirror-image falsely detected obstacles under extremely close conditions, so as to avoid excessive avoidance and track oscillation caused by the virtual image of mirror reflection, thereby improving the flight stability and safety of the aircraft.

[0080] To facilitate better implementation of the above obstacle recognition method, this application also provides a corresponding obstacle recognition device. The meanings of the terms used are the same as in the above obstacle recognition method; for specific implementation details, please refer to the descriptions in the above method embodiments.

[0081] Please refer to Figure 7 , Figure 7 This is a schematic diagram of the obstacle recognition device provided in an embodiment of this application, as shown below. Figure 7 As shown, the obstacle recognition device may include an occupancy prediction module 210, a region generation module 220, a first filtering module 230, a second filtering module 240, a state acquisition module 250, a mirror matching module 260, and a matching verification module 270, wherein... Occupation prediction module 210 is used to acquire the current circumferential image of the aircraft, perform three-dimensional occupancy prediction on the current circumferential image, and obtain the voxel occupancy distribution of the flight space where the aircraft is located. The region generation module 220 is used to extract building surface voxels from the voxel occupancy distribution and generate candidate reflection regions based on the building surface voxels. The first screening module 230 is used to identify candidate obstacles in the flight space based on the voxel occupancy distribution and mark candidate obstacles located outside the candidate reflection area as real obstacles. The second screening module 240, for each candidate obstacle located within the candidate reflection area, extracts the occlusion voxel between the aircraft and the candidate obstacle from the voxel occupancy distribution, and identifies whether the candidate obstacle is occluded based on the occlusion voxel; if the candidate obstacle is identified as occluded, the candidate obstacle is marked as a mirror false detection candidate obstacle, otherwise it is marked as a real obstacle. The state acquisition module 250 is used to acquire multi-dimensional state information of the aircraft and multi-dimensional state information of real obstacles; The mirror matching module 260, for each mirror false detection candidate obstacle, acquires the multi-dimensional state information of the mirror false detection candidate obstacle, and calculates the multi-dimensional state information of the assumed real obstacle corresponding to the mirror false detection candidate obstacle based on the candidate reflection area corresponding to the mirror false detection candidate obstacle; and establishes a one-to-one binding relationship between the mirror false detection candidate obstacle and the aircraft or real obstacle based on the multi-dimensional state information of the assumed real obstacle, the multi-dimensional state information of the aircraft, and the multi-dimensional state information of the real obstacle. The matching and verification module 270 verifies whether the one-to-one binding relationship corresponding to each mirror false detection candidate obstacle is continuously established within the time window of the current circumferential image for each mirror false detection candidate obstacle. If it is continuously established, the mirror false detection candidate obstacle is determined to be a mirror false detection obstacle.

[0082] Optionally, in one embodiment, the region generation module 220 is used to extract the building surface contour based on the building surface voxels, and to perform plane fitting based on the center point of the building surface voxels to obtain a fitting plane; and to generate candidate reflection regions based on the building surface contour and the fitting plane.

[0083] Optionally, in one embodiment, the region generation module 220 is further configured to obtain the fitting residual of the fitting plane; if the fitting residual is less than the residual threshold, then a candidate reflection region is generated based on the building surface contour and the fitting plane.

[0084] Optionally, in one embodiment, the second screening module 240 is used to obtain the three-dimensional bounding box of the candidate obstacle, construct the observation cone of the aircraft to the candidate obstacle based on the three-dimensional bounding box, extract the occlusion voxels located in the observation cone from the voxel occupancy distribution, and identify whether the candidate obstacle is occluded based on the occlusion voxels.

[0085] Optionally, in one embodiment, the second screening module 240 is used to sample multiple observation line segments from the aircraft to the candidate obstacle within the observation cone; determine the proportion of the number of line segments that pass through the occlusion voxel among all observation line segments; if the proportion of the number of line segments is greater than or equal to the proportion threshold, it is determined that the candidate obstacle is occluded; otherwise, it is determined that the candidate obstacle is not occluded.

[0086] Optionally, in one embodiment, the mirror matching module 260 is used to obtain the category difference, position difference, speed difference, and orientation difference between the assumed real obstacle and the aircraft based on the multi-dimensional state information of the assumed real obstacle and the multi-dimensional state information of the aircraft, and to fuse the category difference, position difference, speed difference, and orientation difference between the assumed real obstacle and the aircraft to obtain the comprehensive difference between the assumed real obstacle and the aircraft; to obtain the category difference, position difference, speed difference, and orientation difference between the assumed real obstacle and the real obstacle based on the multi-dimensional state information of the assumed real obstacle and the multi-dimensional state information of the real obstacle, and to fuse the category difference, position difference, speed difference, and orientation difference between the assumed real obstacle and the real obstacle to obtain the comprehensive difference between the assumed real obstacle and the real obstacle; and to establish a one-to-one binding relationship between the mirror false detection candidate obstacle and the aircraft or the real obstacle based on the comprehensive difference between the assumed real obstacle and the aircraft and the real obstacle.

[0087] Optionally, in one embodiment, the mirror matching module 260 is used to perform a weighted summation of the category differences, position differences, speed differences, and orientation differences between the assumed real obstacle and the aircraft to obtain the comprehensive differences between the assumed real obstacle and the aircraft; wherein, the weight coefficient of each difference item is determined according to the importance of its data dimension in obstacle avoidance decision-making.

[0088] Optionally, in one embodiment, the mirror matching module 260 is further configured to mark the mirror misdetected candidate obstacle as a real obstacle when it fails to establish a one-to-one binding relationship between the mirror misdetected candidate obstacle and the aircraft or the real obstacle based on the comprehensive differences between the assumed real obstacle and the aircraft and the real obstacle.

[0089] Optionally, in one embodiment, the obstacle recognition device provided in this application further includes an obstacle avoidance optimization module, which is used to reduce the obstacle avoidance priority of falsely detected obstacles in obstacle avoidance decision-making.

[0090] It should be noted that the information interaction and execution process between the above modules are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0091] In one embodiment, an aircraft is provided whose internal structure can be as follows: Figure 8As shown, the aircraft includes a main body and a memory 310, a processor 320, a power supply 330, a sensor 340, a communication module 350, a positioning module 360, a drive mechanism 370, and a bus 380, all mounted on the main body. The processor 320 is coupled to the memory 310, power supply 330, sensor 340, communication module 350, positioning module 360, and drive mechanism 370 via the bus 380.

[0092] Memory 310 may include one or more random access memories (RAM) and one or more non-volatile memories (NVM). The RAM can be directly read and written by the processor 320 and can be used to store executable programs (such as machine instructions) of the operating system or other running programs, as well as user and application data. The RAM may include static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), and double data rate synchronous dynamic random access memory (DDRAM). (memory, DDR SDRAM, etc.)

[0093] Non-volatile memory can also store executable programs and user and application data, and can be pre-loaded into random access memory for direct reading and writing by the processor 320. Non-volatile memory can include disk storage and flash memory.

[0094] The memory 310 is used to store one or more computer programs. The one or more computer programs are configured to be executed by the processor 320. The one or more computer programs include multiple instructions, which, when executed by the processor 320, can implement the obstacle recognition method provided in this application.

[0095] In other embodiments, the aircraft also includes an external memory interface for connecting to an external memory to expand the aircraft's storage capacity.

[0096] Processor 320 may include one or more processing units, such as: application processor (AP), modem processor, graphics processing unit (GPU), image signal processor (ISP), controller, video codec, digital signal processor (DSP), baseband processor, and / or neural network processor. Processing units (NPUs), etc. Different processing units can be independent devices or integrated into one or more processors.

[0097] The processor 320 provides computing and control capabilities, such as executing computer programs stored in the memory 310 to implement the obstacle recognition method provided in this application.

[0098] Power source 330 is used to supply power to the aircraft. In one embodiment of this application, power source 330 may include any one or more power supply devices such as batteries, fuel generators, and solar power modules.

[0099] Sensor 340 is used to acquire information for the aircraft, such as environmental information and aircraft movement information. In one embodiment of this application, sensor 340 may include a surround-view camera, and may also include one or more sensors of the types such as lidar, millimeter-wave radar, and infrared sensors.

[0100] The communication module 350 is used to enable communication between the aircraft and other devices. In one embodiment of this application, the communication module 350 can interact with other devices via wired and / or wireless communication. The aforementioned wireless communication may include one or more combinations of communication methods such as Bluetooth communication, Wi-Fi communication, and Near Field Communication (NFC).

[0101] The positioning module 360 ​​is used to determine the position of the aircraft. In some embodiments of this application, the positioning module 360 ​​may include one or more of the following types of positioning modules: Global Navigation Satellite System (GNSS), Inertial Navigation System, Real-time Kinematic (RTK) Carrier Phase Differential System, etc.

[0102] The drive mechanism 370 is used to drive the aircraft to fly, and can be different types of drive mechanisms such as jet engines, propeller engines, and rotors. In some embodiments of this application, the drive mechanism 370 can realize the flight function of the aircraft according to the control of the processor 320.

[0103] Bus 380 is used at least to provide a channel for communication between the memory 310, processor 320, power supply 330, sensor 340, communication module 350, positioning module 360, and drive mechanism 370 in the aircraft.

[0104] The main body of the aircraft can be a cargo hold or a crew cabin, used to carry goods or personnel.

[0105] It is understood that the structures illustrated in the embodiments of this application do not constitute a specific limitation on the aircraft. In other embodiments of this application, the aircraft may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0106] For example, please refer to Figure 9 The actual product form of the aircraft can be an electric vertical take-off and landing aircraft, with a compound rotor as its drive mechanism and a crew cabin for carrying personnel as its main body.

[0107] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0108] This application also provides a computer program product, which includes a computer program that, when executed on a processor, causes the processor to implement the steps in the obstacle recognition method provided in this application.

[0109] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0110] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application 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 this application, and should all be included within the protection scope of this application.

[0111] It should be noted that when the above embodiments of this application are applied to specific products or technologies, user-related data is involved, and user permission or consent is required. Furthermore, the collection, use, and processing of such data must comply with the relevant laws, regulations, and standards of the relevant countries and regions.

Claims

1. An obstacle recognition method, characterized in that, include: The current circumferential image of the aircraft is acquired, and a three-dimensional occupancy prediction is performed on the current circumferential image to obtain the voxel occupancy distribution of the flight space in which the aircraft is located. Voxels on the building surface are extracted from the voxel occupancy distribution, and candidate reflection regions are generated based on the voxels on the building surface. Candidate obstacles in the flight space are identified based on the voxel occupancy distribution, and candidate obstacles located outside the candidate reflection area are marked as real obstacles; For each candidate obstacle located within the candidate reflection region, an occlusion voxel located between the aircraft and the candidate obstacle is extracted from the voxel occupancy distribution, and the candidate obstacle is identified as occluded based on the occlusion voxel. If the candidate obstacle is detected to be occluded, it is marked as a mirrored false detection candidate obstacle; otherwise, it is marked as a real obstacle. Obtain multi-dimensional state information of the aircraft and multi-dimensional state information of the real obstacles; For each mirrored false detection candidate obstacle, obtain the multi-dimensional state information of the mirrored false detection candidate obstacle, and calculate the multi-dimensional state information of the assumed real obstacle corresponding to the mirrored false detection candidate obstacle based on the candidate reflection area corresponding to the mirrored false detection candidate obstacle. Based on the multidimensional state information of the assumed real obstacle, the multidimensional state information of the aircraft, and the multidimensional state information of the real obstacle, a one-to-one binding relationship is established between the mirrored false detection candidate obstacle and the aircraft or the real obstacle. For each mirror false detection candidate obstacle, verify whether the one-to-one binding relationship corresponding to the mirror false detection candidate obstacle is continuously established within the time window of the current circumferential image; If the condition persists, the mirrored false detection candidate obstacle will be classified as a mirrored false detection obstacle.

2. The obstacle recognition method according to claim 1, characterized in that, The step of generating candidate reflection regions based on the voxels of the building surface includes: The building surface contour is extracted based on the building surface voxels, and a plane is fitted based on the center points of the building surface voxels to obtain the fitted plane; The candidate reflection region is generated based on the building surface profile and the fitted plane.

3. The obstacle recognition method according to claim 2, characterized in that, Before generating the candidate reflection region based on the building surface contour and the fitted plane, the method further includes: Obtain the fitting residual of the fitted plane; The step of generating the candidate reflection region based on the building surface contour and the fitted plane includes: If the fitting residual is less than the residual threshold, then the candidate reflection region is generated based on the building surface profile and the fitting plane.

4. The obstacle recognition method according to claim 1, characterized in that, The step of extracting the occlusion voxels located between the aircraft and the candidate obstacle from the voxel occupancy distribution, and identifying whether the candidate obstacle is occluded based on the occlusion voxels, includes: Obtain the 3D bounding box of the candidate obstacle, and construct the observation frustum of the aircraft to the candidate obstacle based on the 3D bounding box; Extract the occlusion voxels located within the observation cone from the voxel occupancy distribution, and identify whether the candidate obstacle is occluded based on the occlusion voxels.

5. The obstacle recognition method according to claim 4, characterized in that, The step of identifying whether the candidate obstacle is occluded based on the occlusion voxel includes: Multiple observation line segments from the aircraft to the candidate obstacle are sampled within the observation cone; Determine the percentage of line segments that pass through the occlusion voxel among all observed line segments; if the percentage of line segments is greater than or equal to the percentage threshold, then the candidate obstacle is determined to be occluded; otherwise, the candidate obstacle is determined to be unoccluded.

6. The obstacle recognition method according to claim 1, characterized in that, The step of establishing a one-to-one binding relationship between the mirrored false detection candidate obstacle and the aircraft or the real obstacle based on the multidimensional state information of the assumed real obstacle, the multidimensional state information of the aircraft, and the multidimensional state information of the real obstacle includes: Based on the multidimensional state information of the assumed real obstacle and the multidimensional state information of the aircraft, the category difference, position difference, speed difference and orientation difference between the assumed real obstacle and the aircraft are obtained, and the category difference, position difference, speed difference and orientation difference between the assumed real obstacle and the aircraft are fused to obtain the comprehensive difference between the assumed real obstacle and the aircraft. Based on the multidimensional state information of the assumed real obstacle and the real obstacle, the category difference, position difference, speed difference and orientation difference between the assumed real obstacle and the real obstacle are obtained, and the category difference, position difference, speed difference and orientation difference between the assumed real obstacle and the real obstacle are fused to obtain the comprehensive difference between the assumed real obstacle and the real obstacle. Based on the combined differences between the assumed real obstacle and the aircraft and the real obstacle, a one-to-one binding relationship is established between the mirrored false detection candidate obstacle and the aircraft or the real obstacle.

7. The obstacle recognition method according to claim 6, characterized in that, The fusion of category differences, position differences, speed differences, and orientation differences between the assumed real obstacle and the aircraft yields a comprehensive difference between the assumed real obstacle and the aircraft, including: The weighted summation of the category differences, position differences, speed differences, and orientation differences between the assumed real obstacle and the aircraft yields the comprehensive difference between the assumed real obstacle and the aircraft; wherein, the weight coefficient of each difference item is determined according to the importance of its data dimension in obstacle avoidance decision-making.

8. The obstacle recognition method according to claim 6, characterized in that, Also includes: If it fails to establish a one-to-one binding relationship between the mirrored false detection candidate obstacle and the aircraft or the real obstacle based on the comprehensive differences between the assumed real obstacle and the aircraft and the real obstacle, then the mirrored false detection candidate obstacle is marked as a real obstacle.

9. The obstacle recognition method according to any one of claims 1-8, characterized in that, After determining the mirrored false detection candidate obstacle as a mirrored false detection obstacle, the process further includes: Reduce the obstacle avoidance priority of the falsely detected obstacles in the obstacle avoidance decision.

10. An aircraft, characterized in that, include: ontology; A drive mechanism is used to drive the body to fly; A memory, located in the main body, is used to store computer programs; A processor, disposed in the body, is used to execute the computer program to implement the obstacle recognition method according to any one of claims 1 to 9.

Images