Target detection method, device, equipment, storage medium and program product

Abstract

The application relates to a target detection method, device, equipment, storage medium and program product. The method comprises the following steps: acquiring a power distribution network operation image shot in a power distribution network inspection process; inputting the power distribution network operation image into a perception algorithm model, and acquiring a first perception result output by the perception algorithm model; the first perception result comprises first information of a target object with hidden dangers in the power distribution network operation image; the perception algorithm model comprises a shared feature extraction backbone taking a depth separable convolution as a core and integrating a convolution block attention unit; first point cloud data and three-dimensional coordinates of the target object are sent to a server, and a second perception result sent by the server is received, the second perception result comprising second information of the target object, and the perception accuracy of the second information being higher than that of the first information. The method can improve the image recognition speed in the power distribution network operation.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a target detection method, apparatus, device, storage medium, and program product. Background Technology

[0002] With the increasing scale of power distribution networks, manual inspection methods face problems such as low efficiency, significant safety hazards, and inaccurate data recording. Applying augmented reality (AR) and artificial intelligence (AI) technologies to the field of power distribution network operations, by overlaying equipment information and operation instructions in real time in the field of view of operators and performing intelligent detection, has become an important direction for improving the level of intelligence in operations.

[0003] In related technologies, during power distribution network operations, the main reliance is on two-dimensional target detection algorithms on the server to identify enhanced information such as category information and location information of various devices in the power distribution network system, output rectangular bounding boxes of various devices, and send the output rectangular bounding boxes of various devices to the terminal.

[0004] However, due to the reliance on the identification results transmitted back by the wireless server, there is a significant communication delay, resulting in a slow image recognition speed in power distribution network operations, which cannot meet the millisecond-level real-time identification requirements for live parts in power distribution network operations. Summary of the Invention

[0005] Therefore, it is necessary to provide a target detection method, apparatus, equipment, storage medium, and program product that can improve the image recognition speed in power distribution network operations, addressing the aforementioned technical problems.

[0006] Firstly, this application provides a target detection method applied to a terminal, comprising:

[0007] Acquire images of power distribution network operations captured during power distribution network inspections;

[0008] The power distribution network operation image is input into the perception algorithm model, and the first perception result output by the perception algorithm model is obtained; the first perception result includes the first information of the target object with potential hazards in the power distribution network operation image; the perception algorithm model includes a shared feature extraction backbone with depthwise separable convolution as the core and integrated convolutional block attention units;

[0009] The first point cloud data and three-dimensional coordinates of the target object are sent to the server, and the second perception result sent by the server is received. The second perception result includes second information of the target object, and the perception accuracy of the second information is higher than that of the first information.

[0010] In one embodiment, before sending the first point cloud data and three-dimensional coordinates of the target object to the server, the method further includes:

[0011] Determine the calculated parallax of the binocular camera in the terminal;

[0012] The depth of the target object in the coordinate system of the binocular camera is determined based on the calculated parallax, baseline distance, and effective focal length of the binocular camera.

[0013] The three-dimensional coordinates of the target object are determined based on the depth of the target object in the coordinate system of the stereo camera.

[0014] In one embodiment, after receiving the second sensing result sent by the server, the method further includes:

[0015] The initial pose of the terminal is determined by tracking the movement of target feature points in the power distribution operation image between two adjacent frames using a streaming light algorithm.

[0016] Based on reference points with known physical characteristics, efficient perspective point pose calculation is performed to determine the real-time pose of the terminal relative to the physical environment.

[0017] The current pose of the terminal is determined based on the preliminary pose and the real-time pose.

[0018] Based on the current pose of the terminal, the second information of the target object is displayed in the display component of the terminal.

[0019] In one embodiment, after determining the current pose of the terminal based on the preliminary pose and the real-time pose, the method further includes:

[0020] Based on the environmental information of the terminal, the current pose of the terminal, and the preset safe distance between the target object, a safe distance threshold between the terminal and the target object is determined;

[0021] Based on the current ranging mode of the terminal, determine the current distance between the terminal and the target object;

[0022] Based on the safe distance threshold of the target object and the current distance between the terminal and the target object, a safety warning message is displayed in the display component of the terminal.

[0023] In one embodiment, after receiving the second sensing result sent by the server, the method further includes:

[0024] By using spatial anchor points, the device history parameters of the target object are bound to the three-dimensional coordinates of the target object;

[0025] In response to a user's gesture triggering operation on the three-dimensional coordinates of the target object, the device history parameters of the target object are displayed in the display component of the terminal.

[0026] Secondly, this application provides a target detection method applied to a server, comprising:

[0027] The receiving terminal sends the first point cloud data and three-dimensional coordinates of the target object, wherein the target object is an object with potential hazards in the power distribution operation image identified by the terminal;

[0028] Spatially align the first point cloud data to obtain the second point cloud data;

[0029] By transposing sparse convolution, the second point cloud data at different resolutions are weighted and fused to obtain the third point cloud data;

[0030] Using the three-dimensional coordinates as a priori position, the third point cloud data is identified to obtain a second perception result. The second perception result includes second information of the target object. The perception accuracy of the second information is higher than that of the first information. The first information is the information of the target object identified by the terminal.

[0031] The second sensing result is sent to the terminal.

[0032] Thirdly, this application also provides a target detection device, comprising:

[0033] The image acquisition module is used to acquire images of power distribution network operations captured during power distribution network inspections.

[0034] The first perception module is used to input the power distribution network operation image into the perception algorithm model and obtain the first perception result output by the perception algorithm model; the first perception result includes the first information of the target object with potential hazards in the power distribution network operation image; the perception algorithm model includes a shared feature extraction backbone with depthwise separable convolution as the core and integrated convolutional block attention units;

[0035] The first sending module is used to send the first point cloud data and three-dimensional coordinates of the target object to the server.

[0036] The first receiving module is used to receive the second perception result sent by the server. The second perception result includes second information of the target object, and the perception accuracy of the second information is higher than that of the first information.

[0037] In one embodiment, the first sensing module is further configured to determine the calculated parallax of the binocular camera of the terminal; determine the depth of the target object in the coordinate system of the binocular camera based on the calculated parallax, baseline distance and effective focal length of the binocular camera; and determine the three-dimensional coordinates of the target object based on the depth of the target object in the coordinate system of the binocular camera.

[0038] In one embodiment, the target detection device further includes:

[0039] The display module is used to track the movement of target feature points in the power distribution operation image between two adjacent frames using a streaming light algorithm to determine the initial pose of the terminal; perform efficient perspective point pose calculation based on reference points with known physical characteristics to determine the real-time pose of the terminal relative to the physical environment; determine the current pose of the terminal based on the initial pose and the real-time pose; and display the second information of the target object in the display component of the terminal based on the current pose of the terminal.

[0040] In one embodiment, the display module is further configured to determine a safety distance threshold between the terminal and the target object based on the environmental information where the terminal is located, the current pose of the terminal, and a preset safety distance between the target object; determine the current distance between the terminal and the target object based on the current ranging mode of the terminal; and display a safety prompt message in the display component of the terminal based on the safety distance threshold of the target object and the current distance between the terminal and the target object.

[0041] In one embodiment, the display module is further configured to bind the device history parameters of the target object to the three-dimensional coordinates of the target object via spatial anchor points; and to display the device history parameters of the target object in the display component of the terminal in response to a user's gesture triggering operation on the three-dimensional coordinates of the target object.

[0042] Fourthly, this application provides a target detection device applied to a server, comprising:

[0043] The second receiving module is used to receive the first point cloud data and three-dimensional coordinates of the target object sent by the terminal. The target object is an object with potential hazards in the power distribution operation image identified by the terminal.

[0044] The point cloud processing module is used to spatially align the first point cloud data to obtain the second point cloud data; and to perform weighted fusion of the second point cloud data at different resolutions through transposed sparse convolution to obtain the third point cloud data.

[0045] The second recognition module is used to recognize the third point cloud data with the three-dimensional coordinates as the position prior to obtain a second perception result. The second perception result includes the second information of the target object. The perception accuracy of the second information is higher than that of the first information. The first information is the information of the target object recognized by the terminal.

[0046] The second sending module is used to send the second sensing result to the terminal.

[0047] Fifthly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the target detection method of the first or second aspect described above.

[0048] Sixthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the target detection method of the first or second aspect described above.

[0049] In a seventh aspect, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the target detection method described in the first or second aspect above.

[0050] The aforementioned target detection method, apparatus, equipment, storage medium, and program products acquire distribution network operation images captured during power distribution network inspections; input the distribution network operation images into a perception algorithm model and obtain the first perception result output by the perception algorithm model; the first perception result includes first information about target objects with potential hazards in the distribution network operation images; the perception algorithm model includes a shared feature extraction backbone with depthwise separable convolution as its core and integrated convolutional block attention units; the first point cloud data and three-dimensional coordinates of the target object are sent to a server, and a second perception result sent by the server is received; the second perception result includes second information about the target object, and the perception accuracy of the second information is higher than that of the first information. Because the distribution network operation images are first feature extracted on the terminal side using a shared feature extraction backbone with depthwise separable convolution as its core and integrated convolutional block attention units to obtain the first perception result of the target object, and then high-precision recognition is performed on the server, this method does not completely rely on the feedback of perception results from the server. High-precision target object recognition can be performed even under low computing power conditions on the terminal side, thereby improving the image recognition speed in distribution network operations. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 An application environment diagram of a target detection method provided in an embodiment of this application;

[0053] Figure 2 A schematic flowchart of a target detection method provided in an embodiment of this application;

[0054] Figure 3 A signaling interaction diagram of a target detection method provided in an embodiment of this application;

[0055] Figure 4 A flowchart illustrating another target detection method provided in an embodiment of this application;

[0056] Figure 5 A schematic flowchart illustrating another target detection method provided in an embodiment of this application;

[0057] Figure 6 A flowchart illustrating yet another target detection method provided in an embodiment of this application;

[0058] Figure 7 A structural block diagram of a target detection device provided in an embodiment of this application;

[0059] Figure 8 A structural block diagram of another target detection device provided in the embodiments of this application;

[0060] Figure 9 An internal structural diagram of a computer device provided in an embodiment of this application;

[0061] Figure 10 This is an internal structural diagram of another computer device provided in an embodiment of this application. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of 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.

[0063] The relevant technologies will be explained below.

[0064] With the ever-expanding scale of power distribution networks, traditional manual inspection and operation methods face problems such as low efficiency, significant safety hazards, and inaccurate data recording. Applying augmented reality (AR) and artificial intelligence (AI) technologies to the field of power distribution network operations, by overlaying equipment information and operation instructions in real time in the operator's field of vision and performing intelligent detection, has become an important direction for improving the level of intelligence in operations. However, existing technologies still have shortcomings in terms of target perception dimension, real-time computing, and accuracy of spatial interaction.

[0065] First, in related technologies, the target recognition, ranging, and spatial positioning functions are not well coupled with the front end of the head-mounted AR glasses. The 3D detection and ranging tasks often need to be deployed on the back-end server, relying on the wireless network to transmit the perception results of the server back to the terminal. This results in significant communication delays, leading to slow image recognition speed in power distribution network operations, which cannot meet the millisecond-level real-time recognition requirements of live parts in power distribution network operations.

[0066] Secondly, during power distribution network operations, the two-dimensional target detection algorithm on the server identifies enhanced information such as the category and location information of various devices in the power distribution network system, and outputs rectangular bounding boxes for each device. However, these rectangular bounding boxes are based on device-level identification and cannot achieve pixel-level instance segmentation. This results in the enhanced information being only generally superimposed on the target area, making it difficult to provide fine guidance for specific contours of insulators, clamps, etc., to mark damaged areas, or to measure surface dimensions, thus limiting its applicability in complex power distribution network operations.

[0067] Furthermore, the relevant technologies have not fully integrated the Simultaneous Localization and Mapping (SLAM) data of the AR glasses themselves. They still need to rely on preset spatial anchor points or offline camera calibration for virtual information registration, which makes the initialization process cumbersome and prone to virtual information drift in dynamic occlusion or sudden changes in lighting.

[0068] Finally, the related technologies suffer from high structural complexity in path interaction and safety early warning. Some solutions require additional binocular camera modules or drone platforms, resulting in low system integration, high hardware costs, and difficulty in achieving seamless integrated operation of detection, segmentation, and ranging functions on lightweight head-mounted devices.

[0069] To address the aforementioned issues, embodiments of this application provide a target detection method, apparatus, device, storage medium, and program product. On the terminal side, a shared feature extraction backbone with depthwise separable convolution as its core and integrating convolutional block attention units is first used to extract features from the power distribution network operation image to obtain the first perception result of the target object. Then, high-precision recognition is performed on the server, thereby enabling high-precision target object recognition even under low computing power conditions on the terminal side without completely relying on the feedback of perception results from the server, thus improving the image recognition speed in power distribution network operations.

[0070] The application scenarios of the target detection method provided in this application are described below.

[0071] The target detection method provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104, or it can be located in the cloud or on another network server.

[0072] Terminal 102 first acquires images of power distribution network operations captured during power distribution network inspections. Then, terminal 102 inputs the images into a perception algorithm model and obtains the first perception result output by the model. The first perception result includes first information about target objects with potential hazards in the power distribution network operation images. The perception algorithm model includes a shared feature extraction backbone with depthwise separable convolution as its core and integrated convolutional block attention units. Finally, terminal 102 sends the first point cloud data and three-dimensional coordinates of the target objects to server 104 and receives the second perception result from server 104. The second perception result includes second information about the target objects, and the perception accuracy of the second information is higher than that of the first information.

[0073] The terminal 102 can be, but is not limited to, various head-mounted devices, including virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. The server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0074] In one exemplary embodiment, such as Figure 2 As shown, a target detection method is provided, which is applied to... Figure 1 The following explanation uses the terminal as an example, including S201 to S203. Among them:

[0075] S201. Acquire images of power distribution network operations captured during power distribution network inspection.

[0076] In this application, the user can wear the aforementioned terminal during the power grid inspection process, thereby the terminal can acquire images of the power distribution network operation captured during the power distribution network inspection.

[0077] The aforementioned terminals may include, but are not limited to, various head-mounted devices, including VR devices, AR devices, smart glasses, etc. It should be noted that these terminals may be equipped with cameras, such as binocular cameras, to capture images of power distribution operations.

[0078] It should be noted that the embodiments of this application do not limit the images of power distribution network operations. They can be any images during the power distribution network inspection process, including images of the power distribution network operation environment, images of power equipment in normal condition, or images of power equipment in abnormal condition. Among them, power equipment in abnormal condition can include power equipment with foreign objects attached, power equipment with abnormal emission deformation, etc.

[0079] For example, the above-mentioned power distribution network operation images can be real-time images or images captured after the user triggers a shooting command; this application embodiment does not limit this.

[0080] S202. Input the distribution network operation image into the sensing algorithm model and obtain the first sensing result output by the sensing algorithm model.

[0081] The perception algorithm model will be explained below.

[0082] In some embodiments, the perceptual algorithm model includes a shared feature extraction backbone with depthwise separable convolution as its core and integrated convolutional block attention units.

[0083] For example, the aforementioned perception algorithm model can use the reconstructed YOLOv5s as its basic architecture. To reduce computational overhead, the backbone network can adopt the MobileNetV3 structure, using depthwise separable convolutions instead of traditional convolutions.

[0084] For example, if the feature map dimension corresponding to the power distribution operation image input to the backbone network is the power distribution operation image, and the convolution kernel size is... The output channels are N. Depthwise separable convolution decomposes the task into channel-wise convolution and pointwise convolution, with the following total computational cost. Compared to standard convolution, its computational cost is significantly reduced, with the computational cost ratio being approximately [missing information]. This allows the terminal to maintain a high frame rate detection even with limited computing power.

[0085]

[0086] in, The convolution kernel spatial size, Number of output channels The number of feature map channels. The spatial dimensions of the input feature map.

[0087] For example, the shared feature extraction backbone can also integrate convolutional block attention units to improve the recognition accuracy of small foreign objects (e.g., kite strings, plastic films). The convolutional block attention unit can include channel attention and spatial attention. Channel attention... and spatial attention By combining convolutional blocks in series, the attention unit can enhance the weights of useful features and suppress background noise. The mathematical expression for the attention unit is as follows:

[0088]

[0089] in, This indicates element-wise multiplication; For channel attention, For spatial attention, Input feature map; Channel attention-weighted feature maps; This is the final output feature map.

[0090] It should be understood that the embodiments of this application, through the convolutional block attention unit, can help to more accurately locate the target region of interest (ROI) (i.e., the target object with potential hazards) in the complex power grid background of the distribution network operation image.

[0091] In some embodiments, the perception algorithm model may further include a target detection branch and an instance segmentation branch. The target detection branch can extract the feature map output by the backbone based on shared features to perform category recognition and bounding box localization of target objects in the distribution network operation image. The instance segmentation branch can extract the feature map output by the backbone based on shared features to perform pixel-level contour segmentation of the distribution network operation image, and finally obtain the first perception result of integrated synchronous output.

[0092] The target object is the object with potential hazards in the power distribution network operation image identified by the terminal. The first perception result includes first information about the target object with potential hazards in the power distribution network operation image. The first information may include the category information, bounding box, and pixel-level contour segmentation information of the target object identified by the perception algorithm model.

[0093] In this application, a terminal-side perception algorithm model is used to identify distribution network operation images. Compared to the two-dimensional detection that relies on backend servers in related technologies, this method can provide pixel-level target region descriptions for subsequent accurate enhancement rendering under low computing power conditions on the terminal side. Furthermore, it does not rely on backend servers or offline calibration, thus avoiding problems such as response lag and complex initialization, thereby improving the image recognition speed in distribution network operations.

[0094] In some embodiments, after obtaining the first perception result output by the perception algorithm model, the terminal can also use a binocular camera to perform preliminary spatial positioning and determine the three-dimensional coordinates of the target object.

[0095] For example, the terminal can first determine the calculated parallax of its binocular camera. Then, based on the calculated parallax, baseline distance, and effective focal length of the binocular camera, the terminal determines the depth of the target object in the binocular camera's coordinate system. Finally, based on the depth of the target object in the binocular camera's coordinate system, the terminal determines the three-dimensional coordinates of the target object.

[0096] For example, the terminal can calculate the disparity by feature matching of the left and right images from a binocular camera. Subsequently, based on the camera baseline distance... And effective focal length is The depth of the target object in the coordinate system of the stereo camera is calculated using the principles of perspective geometry. Among them, depth The calculation formula is as follows:

[0097]

[0098] For example, calculating the depth of the target object in the coordinate system of the stereo camera. Then, the coordinates of the image center point can be combined. and intrinsic parameters This allows us to obtain the three-dimensional coordinates of the target object on the terminal side, which can then be used as a location prior for high-precision reconstruction on the server side.

[0099] S203. Send the first point cloud data and three-dimensional coordinates of the target object to the server, and receive the second perception result sent by the server.

[0100] The first point cloud data consists of raw point cloud data related to the target object collected by the terminal's sensors. The second perception result includes second information about the target object, with higher perception precision than the first information. For example, the second information may include the target object's category information, bounding box, and pixel-level contour segmentation information (instance segmentation mask) identified by the model on the server side.

[0101] In some embodiments, after receiving the first point cloud data and 3D coordinates sent by the terminal device, the server can spatially align the first point cloud data to obtain second point cloud data. Next, the server performs weighted fusion of the second point cloud data at different resolutions using transposed sparse convolution to obtain third point cloud data. Then, the server uses the 3D coordinates as a position prior to identify the third point cloud data, obtaining a second perception result. Finally, the server sends the second perception result to the terminal.

[0102] For example, after the terminal identifies a target object through a perception algorithm model, it can lock it as a suspected target and collect high-density point cloud data of the target object through sensors on the terminal. The collected high-density point cloud is then streamed to the server as the first point cloud data. The server can use a fully convolutional anchor-point free 3D detection (e.g., FCAF3D) network model and transposed sparse convolutional units to refine the power data and obtain the second perception result.

[0103] For example, the server can first spatially align the first point cloud data to obtain the second point cloud data. Specifically, the server can use the original, unnecessary first point cloud data... After coordinate transformation matrix Mapping to the world coordinate system yields the second point cloud data. This coordinate transformation matrix consists of a rotation matrix. Translation vector Composition, expressed as follows:

[0104]

[0105] For example, considering the wide range of scales of distribution network equipment, Sparse Adaptive Spatial Feature Fusion (S-ASFF) can be introduced to fuse feature point clouds of different resolutions through transposed sparse convolution. Perform dynamic weighted fusion. The fused [number]th [unit / item / etc.] Layer features This refers to the third point cloud data, and its calculation formula is as follows:

[0106]

[0107] in, The spatial coordinates of a single voxel in the feature map; This is the index of the output feature layer after fusion; , and To perform upsampling or downsampling using transposed sparse convolution, X represents the feature values ​​in the second point cloud data, and Y represents the fused voxel feature values; weight parameters... By using the Softmax function for normalization, the network can automatically adjust the contribution of each layer's features based on the local density and geometry of the point cloud.

[0108] In this application, the target object is finely identified through the model on the server side, and the precise bounding box of the target (center point, size, yaw angle) can be output. It can also achieve pixel-level instance segmentation mask, providing a data foundation for subsequent precise ranging.

[0109] It should be understood that after receiving the second perception result sent by the server, the terminal can also determine the terminal's current pose and display the second information of the target object in the terminal's display component based on the terminal's current pose.

[0110] In some embodiments, the terminal can first use a streaming light algorithm to track the movement of target feature points in a power distribution network operation image between adjacent frames to determine the terminal's preliminary pose. Secondly, the terminal can perform efficient perspective point pose calculation based on reference points with known physical characteristics to determine the terminal's real-time pose relative to the physical environment. Thirdly, the terminal can determine its current pose based on the preliminary and real-time poses. Finally, based on the terminal's current pose, the terminal displays second information about the target object in its display component.

[0111] For example, the terminal can use a pyramid-based optical flow method for feature point tracking. In consecutive frames of images, assuming the pixel brightness corresponding to the target feature point is constant, the basic constraint equations for optical flow are established:

[0112]

[0113] in, Let be the partial derivative of pixel brightness in the horizontal direction of the image. Let be the partial derivative of pixel brightness in the vertical direction of the image. Let be the partial derivative of pixel brightness with respect to time. This represents the horizontal displacement component of the pixel. This represents the vertical displacement component of the pixel.

[0114] For example, after establishing the fundamental constraint equations for optical flow, it is possible to... Within a neighborhood window, overdetermined equations are constructed and the least squares method is applied to obtain the movement of the target feature point between two adjacent frames, i.e., the motion vector in the two-dimensional plane. This allows us to obtain the initial pose of the terminal.

[0115] For example, the terminal can also combine reference points with known physical characteristics (e.g., tower markings, switchgear corners, etc.) to perform efficient perspective point (EPnP) pose calculation, solving for the terminal's real-time pose relative to the physical environment. Specifically, four control points can be selected. , reference point The linear combination of control points is represented as follows. Subsequently, singular value decomposition (SVD) is used to calculate the coordinates of the control points in the camera coordinate system, thereby obtaining the rotation matrix R and the translation vector T, and thus obtaining the real-time pose of the terminal relative to the physical environment.

[0116]

[0117] in, For the first The three-dimensional coordinates of a reference point in the world coordinate system for The three-dimensional coordinates of each control point in the world coordinate system These are the coefficients of the linear combination.

[0118] It should be understood that by fusing the terminal's initial pose with its real-time pose relative to the physical environment, the terminal's current pose can be determined. This pose estimation accuracy can reach the millimeter level, ensuring that relevant information in the terminal's display components can be accurately pinned to the physical device.

[0119] For example, the terminal can use its current pose to perform visual odometry compensation to remove drift or ghosting when displaying second information in the terminal's display components, thereby improving the display effect of the display components.

[0120] In some embodiments, after determining its current pose based on the preliminary pose and the real-time pose, the terminal can further determine a safety distance threshold between itself and the target object based on the environmental information, its current pose, and a preset safety distance from the target object. Subsequently, the terminal determines the current distance between itself and the target object based on its current ranging mode. Finally, the terminal displays a safety warning message on its display component based on the safety distance threshold for the target object and the current distance between itself and the target object.

[0121] It should be understood that this application addresses the differentiated safety distance requirements for different voltage levels (such as 10kV and 35kV) in distribution network operations. The system breaks away from the limitations of traditional fixed distance criteria and introduces a dynamic threshold function based on power operation procedures. This allows us to calculate the safe distance threshold for the target object.

[0122] For example, the terminal can identify the device nameplate of the target object in real time, and then retrieve the voltage level V of the target object's charged body from the cloud device ledger by accessing the identified device identifier. Subsequently, a preset safety distance to the target object is determined based on the voltage level V. Then, the preset safety distance to the target object, the pose positioning standard deviation corresponding to the terminal's current pose, and the environmental correction factor corresponding to the environmental information are respectively input into a dynamic threshold function. This allows for the calculation of the safe distance threshold between the terminal and the target object. Dynamic threshold function The formula is shown below:

[0123]

[0124] in, The safe distance threshold for the target object; The preset safety distance for the target object represents the minimum safety distance required by power grid safety regulations (e.g., 0.7m for 10kV and 1.0m for 35kV). The standard deviation of pose localization is used to compensate for the cumulative error in visual odometry caused by rapid head rotation of the user. It serves as an environmental correction factor, used to monitor humidity, visibility, and air pressure, dynamically adjusting air insulation redundancy to ensure operational safety under extreme weather conditions.

[0125] In some embodiments, the terminal's current ranging mode may include geometric center ranging (i.e., initial ranging mode) and instance-based segmentation masking (…). ) with extremely fine-grained ranging (i.e., fine measurement mode).

[0126] For example, geometric center ranging is mainly used for rapid localization of distant targets. Let the current optical center coordinates of the terminal be... The centroid of the bounding box in the second information of the target object is The current distance between the terminal and the target object. The calculation is as follows:

[0127]

[0128] in, The three-dimensional coordinates of the terminal in the world coordinate system. The three-dimensional coordinates of the target object in the world coordinate system.

[0129] For example, based on The extremely fine-grained ranging, targeting the discharge risk at the most prominent part of the charged body, can traverse the set of segmented mask points in the second information of the target object. Each feature point in Real-time search using k-dimensional tree (KD-Tree) indexing, minimizing the physical distance between the terminal and the target object. :

[0130]

[0131] in, Let be the Euclidean distance norm.

[0132] For example, the terminal can also have integrated sound, light, and visual warning feedback logic. The terminal can perform a millisecond-level comparison between the current distance between the terminal and the target object and the target object's safe distance threshold. Combined with the terminal's current pose, the current distance and warning label are accurately fixed on the physical surface of the target object, solving the problem of virtual information drift.

[0133] For example, when near When (e.g., the difference is less than 1 meter), the display component The covered area can be rendered with a red pulse signal. When Less than or equal to At that time, a holographic warning message immediately pops up in the display component, and a voice alarm is issued to the operator through the terminal. At the same time, a violation warning log with geographical location and timestamp is sent to the server.

[0134] For example, the terminal can also have a built-in time-domain smoothing filter to predict and compensate for temporary obstructions (such as tree branches or utility poles) that may occur during the inspection site, using historical pose data to ensure the continuous smoothness of the current distance between the terminal and the target object and avoid false alarms.

[0135] In this application, a dual-modal spatial ranging engine based on a dynamic safety threshold model is constructed by integrating environmental information, the current pose of the terminal, and the preset safe distance of the target object, thereby realizing timely distance warning in power distribution operations.

[0136] In some embodiments, after receiving the second sensing result sent by the server, the terminal can also bind the device history parameters of the target object to the three-dimensional coordinates of the target object through spatial anchor points. Subsequently, in response to the user's gesture triggering operation on the three-dimensional coordinates of the target object, the device history parameters of the target object are displayed in the display component of the terminal.

[0137] For example, the terminal can bind the historical device parameters of the target object to the three-dimensional coordinates of the target object through spatial anchor points, thereby fixing the historical device parameters of the target object in physical spatial coordinates. These historical device parameters may include device status, the distance between the terminal and the target object, and inspection guide information, etc.

[0138] For example, in terms of human-computer interaction, the terminal can also integrate a gesture tracking algorithm. Users can trigger gestures targeting the three-dimensional coordinates of a target object by pointing from a distance or touching it from a distance, thereby retrieving the target object's historical device parameters and displaying them in the terminal's display component.

[0139] For example, the terminal can also automatically identify the values ​​of physical electricity meters. Through technical logic, when the identified current or voltage values ​​exceed the rated range, a troubleshooting manual will pop up on the terminal's display. This interactive solution makes complex power distribution network operations more visual and data-driven, effectively reducing the risk of misoperation.

[0140] This application achieves millisecond-level charged body distance estimation and WYSIWYG spatial anchoring of virtual information, overcoming the drawbacks of high latency, complex initialization, and easy drift caused by relying on backend server data transmission or offline anchor point calibration. Furthermore, based on the target instance's detection box level, segmentation mask level, and ranging data, it dynamically generates and renders multi-level virtual information, including text labels, outline highlighted borders, virtual ranging, and safety warning surfaces, ensuring precise alignment with the real equipment surface or workspace. This mechanism significantly differs from related technologies that only provide a rough target box overlay interaction, achieving more refined and intuitive work guidance and safety warnings.

[0141] The target detection method provided in this application acquires distribution network operation images captured during power distribution network inspection; inputs the distribution network operation images into a perception algorithm model and obtains a first perception result output by the perception algorithm model; the first perception result includes first information about target objects with potential hazards in the distribution network operation images; the perception algorithm model includes a shared feature extraction backbone with depthwise separable convolution as its core and integrated convolutional block attention units; the first point cloud data and three-dimensional coordinates of the target object are sent to a server, and a second perception result sent by the server is received; the second perception result includes second information about the target object, and the perception accuracy of the second information is higher than that of the first information. Because the distribution network operation images are first feature extracted on the terminal side using a shared feature extraction backbone with depthwise separable convolution as its core and integrated convolutional block attention units to obtain the first perception result of the target object, and then high-precision recognition is performed on the server, this method does not completely rely on the feedback of perception results from the server. High-precision target object recognition can be performed even under low computing power conditions on the terminal side, thereby improving the image recognition speed in distribution network operations.

[0142] In one exemplary embodiment, such as Figure 3 As shown, a target detection method is provided to illustrate the interaction process between a terminal and a server. This target detection method includes steps S301 to S307. Wherein:

[0143] S301. The terminal acquires images of power distribution network operations captured during power distribution network inspection.

[0144] S302. The terminal inputs the distribution network operation image into the sensing algorithm model and obtains the first sensing result output by the sensing algorithm model.

[0145] The first perception result includes the first information of the target object with potential hazards in the power distribution network operation image; the perception algorithm model includes a shared feature extraction backbone with depthwise separable convolution as the core and integrated convolutional block attention units.

[0146] S303, The terminal sends the first point cloud data and three-dimensional coordinates of the target object to the server.

[0147] S304. The server spatially aligns the first point cloud data to obtain the second point cloud data.

[0148] S305. The server uses transposed sparse convolution to perform weighted fusion of second point cloud data at different resolutions to obtain third point cloud data.

[0149] S306. The server uses three-dimensional coordinates as a priori position to identify the third point cloud data and obtain the second perception result.

[0150] The second perception result includes second information about the target object. The perception accuracy of the second information is higher than that of the first information. The first information is the information about the target object identified by the terminal.

[0151] S307. The server sends the second sensing result to the terminal.

[0152] In one exemplary embodiment, such as Figure 4 As shown, a target detection method is provided to illustrate how to determine the current pose of a terminal. This target detection method includes steps S401 to S407. Wherein:

[0153] S401. Acquire images of power distribution network operations captured during power distribution network inspection.

[0154] S402. Input the distribution network operation image into the perception algorithm model and obtain the first perception result output by the perception algorithm model; the first perception result includes the first information of the target object with potential hazards in the distribution network operation image; the perception algorithm model includes a shared feature extraction backbone with depthwise separable convolution as the core and integrated convolutional block attention units.

[0155] S403. Send the first point cloud data and three-dimensional coordinates of the target object to the server, and receive the second perception result sent by the server. The second perception result includes the second information of the target object, and the perception accuracy of the second information is higher than that of the first information.

[0156] S404. The target feature points in the power distribution network operation image are tracked between two adjacent frames using the streaming light algorithm to determine the initial pose of the terminal.

[0157] S405. Based on the reference points with known physical characteristics, perform efficient perspective point pose calculation to determine the real-time pose of the terminal relative to the physical environment.

[0158] S406. Determine the current pose of the terminal based on the preliminary pose and the real-time pose.

[0159] S407. Based on the current pose of the terminal, display the second information of the target object in the display component of the terminal.

[0160] In one exemplary embodiment, such as Figure 5 As shown, a target detection method is provided to illustrate how to determine the current pose of a terminal. This target detection method includes steps S501 to S510. Wherein:

[0161] S501. Acquire images of power distribution network operations captured during power distribution network inspection.

[0162] S502. Input the power distribution network operation image into the perception algorithm model and obtain the first perception result output by the perception algorithm model; the first perception result includes the first information of the target object with potential hazards in the power distribution network operation image; the perception algorithm model includes a shared feature extraction backbone with depthwise separable convolution as the core and integrated convolutional block attention units.

[0163] S503. Send the first point cloud data and three-dimensional coordinates of the target object to the server, and receive the second perception result sent by the server. The second perception result includes the second information of the target object, and the perception accuracy of the second information is higher than that of the first information.

[0164] S504. The target feature points in the power distribution network operation image are tracked between two adjacent frames using the streaming light algorithm to determine the initial pose of the terminal.

[0165] S505. Based on the reference points with known physical characteristics, perform efficient perspective point pose calculation to determine the real-time pose of the terminal relative to the physical environment.

[0166] S506. Determine the current pose of the terminal based on the preliminary pose and the real-time pose.

[0167] S507. Based on the current pose of the terminal, display the second information of the target object in the display component of the terminal.

[0168] S508. Determine the safe distance threshold between the terminal and the target object based on the environmental information of the terminal, the current position of the terminal, and the preset safe distance of the target object.

[0169] S509. Determine the current distance between the terminal and the target object based on the terminal's current ranging mode.

[0170] S510. Based on the safe distance threshold of the target object and the current distance between the terminal and the target object, display a safety prompt message in the display component of the terminal.

[0171] In one exemplary embodiment, such as Figure 6 As shown, a target detection method is provided to illustrate how to bind device history parameters of the target object. This target detection method includes steps S601 to S605. Wherein:

[0172] S601. Acquire images of power distribution network operations captured during power distribution network inspection.

[0173] S602. Input the distribution network operation image into the perception algorithm model and obtain the first perception result output by the perception algorithm model; the first perception result includes the first information of the target object with potential hazards in the distribution network operation image; the perception algorithm model includes a shared feature extraction backbone with depthwise separable convolution as the core and integrated convolutional block attention units.

[0174] S603. Send the first point cloud data and three-dimensional coordinates of the target object to the server, and receive the second perception result sent by the server. The second perception result includes the second information of the target object, and the perception accuracy of the second information is higher than that of the first information.

[0175] S604. Bind the historical parameters of the target object to the three-dimensional coordinates of the target object through spatial anchor points.

[0176] S605, In response to a user's gesture triggering operation on the three-dimensional coordinates of the target object, display the device history parameters of the target object in the display component of the terminal.

[0177] The target detection method provided in this application acquires distribution network operation images captured during power distribution network inspection; inputs the distribution network operation images into a perception algorithm model and obtains a first perception result output by the perception algorithm model; the first perception result includes first information about target objects with potential hazards in the distribution network operation images; the perception algorithm model includes a shared feature extraction backbone with depthwise separable convolution as its core and integrated convolutional block attention units; the first point cloud data and three-dimensional coordinates of the target object are sent to a server, and a second perception result sent by the server is received; the second perception result includes second information about the target object, and the perception accuracy of the second information is higher than that of the first information. Because the distribution network operation images are first feature extracted on the terminal side using a shared feature extraction backbone with depthwise separable convolution as its core and integrated convolutional block attention units to obtain the first perception result of the target object, and then high-precision recognition is performed on the server, this method does not completely rely on the feedback of perception results from the server. High-precision target object recognition can be performed even under low computing power conditions on the terminal side, thereby improving the image recognition speed in distribution network operations.

[0178] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0179] Based on the same inventive concept, this application also provides a target detection apparatus for implementing the target detection method described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more target detection apparatus embodiments provided below can be found in the limitations of the target detection method described above, and will not be repeated here.

[0180] In one exemplary embodiment, such as Figure 7 As shown, a target detection device 700 is provided, including: an image acquisition module 701, a first sensing module 702, a first transmitting module 703, and a first receiving module 704, wherein:

[0181] Image acquisition module 701 is used to acquire images of power distribution network operations captured during power distribution network inspection.

[0182] The first perception module 702 is used to input the power distribution network operation image into the perception algorithm model and obtain the first perception result output by the perception algorithm model; the first perception result includes the first information of the target object with potential hazards in the power distribution network operation image; the perception algorithm model includes a shared feature extraction backbone with depthwise separable convolution as the core and integrated convolutional block attention units;

[0183] The first sending module 703 is used to send the first point cloud data and three-dimensional coordinates of the target object to the server.

[0184] The first receiving module 704 is used to receive the second sensing result sent by the server. The second sensing result includes second information of the target object, and the sensing accuracy of the second information is higher than that of the first information.

[0185] In one embodiment, the first sensing module 702 is further configured to determine the calculated parallax of the binocular camera of the terminal; determine the depth of the target object in the coordinate system of the binocular camera based on the calculated parallax, baseline distance and effective focal length of the binocular camera; and determine the three-dimensional coordinates of the target object based on the depth of the target object in the coordinate system of the binocular camera.

[0186] In one embodiment, the target detection device 700 further includes:

[0187] The display module is used to track the movement of target feature points in the power distribution network operation image between two adjacent frames using a streaming light algorithm to determine the initial pose of the terminal; to perform efficient perspective point pose calculation based on reference points with known physical characteristics to determine the real-time pose of the terminal relative to the physical environment; to determine the current pose of the terminal based on the initial pose and the real-time pose; and to display the second information of the target object in the display component of the terminal based on the current pose of the terminal.

[0188] In one embodiment, the display module is further configured to determine a safe distance threshold between the terminal and the target object based on the environmental information where the terminal is located, the current pose of the terminal, and the preset safe distance between the target object; determine the current distance between the terminal and the target object based on the current ranging mode of the terminal; and display a safety prompt message in the display component of the terminal based on the safe distance threshold of the target object and the current distance between the terminal and the target object.

[0189] In one embodiment, the display module is further configured to bind the device history parameters of the target object to the three-dimensional coordinates of the target object through spatial anchor points; and to display the device history parameters of the target object in the display component of the terminal in response to a user's gesture triggering operation on the three-dimensional coordinates of the target object.

[0190] Each module in the aforementioned target detection device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0191] Based on the same inventive concept, this application also provides a target detection apparatus for implementing the target detection method described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more target detection apparatus embodiments provided below can be found in the limitations of the target detection method described above, and will not be repeated here.

[0192] In one exemplary embodiment, such as Figure 8As shown, a target detection device 800 is provided, including: a second receiving module 801, a point cloud processing module 802, a second identification module 803, and a second transmitting module 804, wherein:

[0193] The second receiving module 801 is used to receive the first point cloud data and three-dimensional coordinates of the target object sent by the terminal. The target object is the object with potential hazards in the power distribution operation image identified by the terminal.

[0194] The point cloud processing module 802 is used to spatially align the first point cloud data to obtain the second point cloud data; and to perform weighted fusion of the second point cloud data at different resolutions through transposed sparse convolution to obtain the third point cloud data.

[0195] The second recognition module 803 is used to recognize the third point cloud data with three-dimensional coordinates as the position prior, and obtain the second perception result. The second perception result includes the second information of the target object. The perception accuracy of the second information is higher than that of the first information. The first information is the information of the target object recognized by the terminal.

[0196] The second sending module 804 is used to send the second sensing result to the terminal.

[0197] Each module in the aforementioned target detection device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0198] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 9 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a target detection method on the server side.

[0199] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 10 As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a target detection method on the terminal side. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0200] Those skilled in the art will understand that Figure 9 and 10 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0201] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the target detection method described above.

[0202] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the target detection method described above.

[0203] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the target detection method described above.

[0204] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0205] 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. The 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, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0206] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0207] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A target detection method, characterized in that, Applied to a terminal, the method includes: Acquire images of power distribution network operations captured during power distribution network inspections; The power distribution network operation image is input into the perception algorithm model, and the first perception result output by the perception algorithm model is obtained; the first perception result includes the first information of the target object with potential hazards in the power distribution network operation image; the perception algorithm model includes a shared feature extraction backbone with depthwise separable convolution as the core and integrated convolutional block attention units; The first point cloud data and three-dimensional coordinates of the target object are sent to the server, and the second perception result sent by the server is received. The second perception result includes second information of the target object, and the perception accuracy of the second information is higher than that of the first information.

2. The method according to claim 1, characterized in that, Before sending the first point cloud data and three-dimensional coordinates of the target object to the server, the method further includes: Determine the calculated parallax of the binocular camera in the terminal; The depth of the target object in the coordinate system of the binocular camera is determined based on the calculated parallax, baseline distance, and effective focal length of the binocular camera. The three-dimensional coordinates of the target object are determined based on the depth of the target object in the coordinate system of the stereo camera.

3. The method according to claim 1, characterized in that, After receiving the second sensing result sent by the server, the method further includes: The initial pose of the terminal is determined by tracking the movement of target feature points in the power distribution operation image between two adjacent frames using a streaming light algorithm. Based on reference points with known physical characteristics, efficient perspective point pose calculation is performed to determine the real-time pose of the terminal relative to the physical environment. The current pose of the terminal is determined based on the preliminary pose and the real-time pose. Based on the current pose of the terminal, the second information of the target object is displayed in the display component of the terminal.

4. The method according to claim 3, characterized in that, After determining the current pose of the terminal based on the preliminary pose and the real-time pose, the method further includes: Based on the environmental information of the terminal, the current pose of the terminal, and the preset safe distance between the target object, a safe distance threshold between the terminal and the target object is determined; Based on the current ranging mode of the terminal, determine the current distance between the terminal and the target object; Based on the safe distance threshold of the target object and the current distance between the terminal and the target object, a safety warning message is displayed in the display component of the terminal.

5. The method according to claim 1, characterized in that, After receiving the second sensing result sent by the server, the method further includes: By using spatial anchor points, the device history parameters of the target object are bound to the three-dimensional coordinates of the target object; In response to a user's gesture triggering operation on the three-dimensional coordinates of the target object, the device history parameters of the target object are displayed in the display component of the terminal.

6. A target detection method, characterized in that, Applied to a server, the method includes: The receiving terminal sends the first point cloud data and three-dimensional coordinates of the target object, wherein the target object is an object with potential hazards in the power distribution operation image identified by the terminal; Spatially align the first point cloud data to obtain the second point cloud data; By transposing sparse convolution, the second point cloud data at different resolutions are weighted and fused to obtain the third point cloud data; Using the three-dimensional coordinates as a priori position, the third point cloud data is identified to obtain a second perception result. The second perception result includes second information of the target object. The perception accuracy of the second information is higher than that of the first information. The first information is the information of the target object identified by the terminal. The second sensing result is sent to the terminal.

7. A target detection device, characterized in that, Applied to a terminal, the device includes: The image acquisition module is used to acquire images of power distribution network operations captured during power distribution network inspections. The first perception module is used to input the power distribution network operation image into the perception algorithm model and obtain the first perception result output by the perception algorithm model; the first perception result includes the first information of the target object with potential hazards in the power distribution network operation image; the perception algorithm model includes a shared feature extraction backbone with depthwise separable convolution as the core and integrated convolutional block attention units; The first sending module is used to send the first point cloud data and three-dimensional coordinates of the target object to the server. The first receiving module is used to receive the second perception result sent by the server. The second perception result includes second information of the target object, and the perception accuracy of the second information is higher than that of the first information.

8. A target detection device, characterized in that, Applied to a server, the device includes: The second receiving module is used to receive the first point cloud data and three-dimensional coordinates of the target object sent by the terminal. The target object is an object with potential hazards in the power distribution operation image identified by the terminal. The point cloud processing module is used to spatially align the first point cloud data to obtain the second point cloud data; and to perform weighted fusion of the second point cloud data at different resolutions through transposed sparse convolution to obtain the third point cloud data. The second recognition module is used to recognize the third point cloud data with the three-dimensional coordinates as the position prior to obtain a second perception result. The second perception result includes the second information of the target object. The perception accuracy of the second information is higher than that of the first information. The first information is the information of the target object recognized by the terminal. The second sending module is used to send the second sensing result to the terminal.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images