UAV power tower detection and autonomous cruising method based on image recognition
Through the drone power tower detection and autonomous cruise method based on image recognition, the convolutional neural network is used to identify the position and distance of the power tower, and the automatic patrol of the drone is realized, which solves the problems of low efficiency and high cost of traditional patrols. Ensure the safety and accuracy of the inspection process.
Patent Information
- Application Number
- CN202110674083.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-06-17
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2041-06-17
AI Technical Summary
The inspection of traditional drone transmission lines is low in efficiency and high in cost, and there are safety hazards in manual operation, which cannot meet the inspection needs of large-scale transmission lines.
The drone power tower detection and autonomous cruise method based on image recognition are adopted to identify the position and distance of the power tower through convolutional neural network, and automatic patrol of the drone is realized, reducing the dependence on pilots and data processing personnel.
It improves the efficiency of drone inspection, reduces costs, ensures the safety and accuracy of the inspection process, and realizes automatic inspection of power poles and towers by drones.
Smart Images

Figure CN113298035B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of automatic inspection of power transmission lines, and in particular relates to an image recognition-based UAV transmission line detection and autonomous cruising method for autonomous inspection of power towers. Background Art
[0002] In recent years, with the rapid development of my country's economy and the gradual improvement of manufacturing capabilities, the country and society's demand for electricity has increased day by day. Against the background of the growing development of large-scale transmission lines such as high voltage, ultra-high voltage and ultra-high voltage, these transmission lines are generally built in sparsely populated areas with complex terrain and harsh environments, which can easily cause aging and damage to equipment, resulting in defects in the transmission capacity of the transmission lines and affecting the normal operation of the transmission equipment. The stability of the power grid directly affects the production and development of society, so the inspection of transmission lines has become particularly important and has been given great attention.
[0003] Traditional manual work is inefficient, has complicated steps, and the personal safety of inspectors cannot be guaranteed. It is no longer suitable for today's large-scale transmission line inspections. The traditional UAV transmission line inspection method mainly uses laser radar, oblique photography and other surveying and mapping methods to obtain high-precision three-dimensional point cloud map information in power towers and line corridors, and offline planning of shooting targets to generate inspection routes. However, due to the high cost of high-precision laser ranging equipment, high maintenance costs, limited effective distance, large weight and high power consumption, it cannot meet the endurance requirements of long-distance inspection operations. At the same time, after the point cloud data is collected, 3D modeling requires a lot of manual participation, large operating workload, time-consuming and labor-intensive, high production costs, and low efficiency.
[0004] At the same time, when manually operating a drone, it is necessary to rely on the pilot and the observer's own vision for observation, which has high requirements for ambient light and weather conditions. Moreover, when the pilot observes the drone on the ground, the observation angle is limited, and the location of some key equipment is easily blocked, which results in the positioning operation of the drone taking a long time. At the same time, due to the safe working distance between power poles and cables, if the distance is misjudged, the drone will be affected by high voltage and be in danger of damage.
[0005] In summary, how to improve the efficiency of autonomous inspections by drones and reduce inspection costs is an urgent problem that technical personnel in this field need to solve. Summary of the invention
[0006] In order to solve the problems existing in the prior art, the present invention provides a UAV power pole tower detection and autonomous cruising method based on image recognition, which is used to realize the automatic inspection operation of the UAV on the power pole tower target, so as to achieve the purpose of improving the inspection efficiency and reducing the cost.
[0007] To achieve the above object, the present invention provides the following technical solutions:
[0008] A method for detecting and autonomously cruising a power pole tower using a drone based on image recognition comprises the following steps:
[0009] S1: manually setting the navigation direction of the drone equipped with the gimbal device, taking a picture of the power transmission line in front of the drone to obtain a first image, extracting the features of the power tower using the constructed convolutional neural network after preprocessing the first image, identifying the positions of all the power towers in the first image and numbering them in sequence;
[0010] S2: The drone flies toward the selected target power pole and records a video, extracts key frame images from the recorded video at a certain time interval, recognizes the key frame images through the convolutional neural network, selects and marks the identified target power pole, and compares them to determine the distance between the drone and the target power pole;
[0011] S3: When the height of the target power pole tower in the key frame image exceeds a certain ratio, the key points on the target power pole tower are identified by the convolutional neural network to determine the angle and distance of the drone relative to the target power pole tower;
[0012] S4: According to the distance and angle data between the drone and the target power tower obtained in step S3, after the drone moves to a direction perpendicular to the transmission line connected to the power tower, the drone is made higher than the target power tower and photographs key points on one side of the target power tower. After photographing one side of the target power tower, the drone adjusts its position to photograph key points on the other side of the target power tower. After that, the drone selects power towers for inspection in sequence according to the numbers;
[0013] S5: Upload the captured image obtained in step S4 to the image library, and add feature information to the captured image.
[0014] Preferably, the convolutional neural network in step S1 for identifying power poles and towers comprises the following steps:
[0015] S100: Obtain a first image and pre-process it;
[0016] S200: the convolutional neural network performs image segmentation on the preprocessed first image, and uses a bounding box to select features of the power pole tower, wherein the parameters contained in the bounding box include a probability and a confidence level of the power pole tower in the bounding box;
[0017] S300: Use the convolutional neural network to output the target detection result, filter out the most optimal frame through non-maximum suppression, and obtain the power tower position.
[0018] Preferably, selecting the power pole tower by the convolutional neural network comprises the following steps:
[0019] S201: Segmenting the preprocessed first image into grids by using the convolutional neural network, wherein the number of grids is set according to the distance between the drone and the power tower;
[0020] S202: distinguishing the shape of the power tower and the surrounding environment, marking the selection frame coordinate information of the power tower, and constructing a power tower image database;
[0021] S203: Using the constructed power pole tower image database to train the convolutional neural network, and obtain key weight information of the convolutional neural network:
[0022] S204: Calibrate the pixel coordinates of the power pole tower in the first image and select the power pole tower by using the trained convolutional neural network.
[0023] Preferably, the convolutional neural network includes a first convolutional layer, a feature extraction layer, and multiple convolutional layers connected in sequence, the feature extraction layer includes a VBL layer, a Resn layer, and a pooling layer, and the VBL layer includes a VGG convolutional layer, a BatchNormalization layer, and a Leaky relu layer.
[0024] Preferably, the pre-training process of the convolutional neural network comprises the following steps:
[0025] S11: Using on-site shooting, shooting the power poles and towers according to different lighting conditions, weather conditions, distances and angles, and calibrating and classifying the power poles and towers images according to the on-site weather and lighting conditions to build a power poles and towers image library;
[0026] S12: Improve the performance of the algorithm by using a variety of data augmentation methods, including horizontal flipping, cropping, zooming in and out, and inputting image data of different sizes and shapes;
[0027] S13: manually mark the power pole tower area, use the labeling tool labeimg to mark the area of interest, and use the CVAT calibration tool to mark the key points on the power pole tower;
[0028] S14: Train the convolutional layer of the convolutional neural network on the constructed power tower image library dataset.
[0029] Preferably, in step S3, when the height of the target power tower in the key frame image exceeds three quarters, the key points on the target power tower are identified by the convolutional neural network to determine the angle and distance of the drone relative to the target power tower.
[0030] Preferably, in step S3, when the drone determines the angle of the target power pole tower, a binocular ranging function carried by the drone is enabled to assist the drone flight.
[0031] Preferably, the distance between the UAV and the target power pole tower is determined by using a gradient search to perform a contour deduction method or a state space modeling method to model the shape and motion state of the target contour.
[0032] Preferably, when the key points on the target power pole tower are identified by the convolutional neural network, the focal length of the pan-tilt device is first reduced to allow the drone to get closer to the target power pole tower, and then the focal length is increased after approaching to a certain distance.
[0033] Compared with the prior art, the present invention has the following beneficial effects:
[0034] The present invention takes a photo of the transmission line, identifies the position of the power tower through an image recognition method based on deep learning, determines the first power tower target to be inspected by the drone, and makes the drone fly to the target power tower. The distance between the drone and the power tower is determined through the video image. When the drone flies close to the power tower, the power tower is identified through a fine recognition mode, and then the angular position of the power tower relative to the drone is determined through the calibration of the key parts of the power tower. The inspection begins based on the determined distance and angle. The above technical solution disclosed in the present application can realize the automatic inspection of power towers by drones without the participation of pilots or data processing personnel, thereby improving the efficiency of circuit inspections and reducing inspection costs. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] Figure 1 A flowchart of a drone detection and autonomous cruising method provided by an embodiment of the present invention.
[0036] Figure 2 A flowchart of a method for using a drone to recognize images of power towers is provided in an embodiment of the present invention.
[0037] Figure 3 A flowchart of a method for using a drone to recognize and select power tower images is provided in an embodiment of the present application.
[0038] Figure 4 A schematic diagram of a neural network structure for drone detection and autonomous cruising provided in an embodiment of the present application. DETAILED DESCRIPTION
[0039] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, the specific implementation methods of the present invention are now described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0040] See attached Figure 1 As shown, it is a flow chart of the provided UAV power tower detection and autonomous cruising method based on image recognition, which mainly includes the following steps:
[0041] S1: Manually set the navigation direction of the drone equipped with the gimbal device, take a high-resolution panoramic photo of the power transmission line in front of the drone, obtain the first image, pre-process the first image, use the constructed convolutional neural network to extract the features of the power tower, identify and select the positions of all power towers in the first image, and number the power towers in sequence according to the pixel size they occupy according to the manually set navigation direction; the high-resolution panoramic photo is used because it is inevitable that trees and hills will appear in the picture when shooting from a long distance. If the image information is not rich enough, it will lead to fewer feature points being extracted, resulting in inaccurate classification during classification. When pre-processing the image, set the threshold according to parameters such as weather and light to reduce environmental noise; considering that the power towers in the transmission line may overlap due to different angles, the navigation direction of the drone when taking off should not be parallel to the direction of the transmission line as much as possible; the number of the power tower identified by the drone can be given manually. If it is not given, the drone will fly to the nearest power tower target in the navigation direction according to the default setting;
[0042] S2: The drone flies to the selected target power pole and records the video. By adjusting the drone, the target power pole is gradually moved to the center of the picture. The video is divided into frames, and key frame images are extracted from the recorded video at a certain time interval. The key frame images are converted into grayscale images to reduce the image data dimension. Then, edge points are extracted and key pixels are retained to minimize the amount of calculation. The key frame images are recognized through a convolutional neural network, and the recognized target power pole is marked and compared. By comparing the power pole contour position of the previous key frame image with the position in the current key frame image, the overlapping part of the target contour of the current key frame image and the previous key frame image is calculated to determine the distance between the drone and the target power pole.
[0043] The method for determining the distance between the UAV and the target power pole tower can be either of the following two methods: using gradient search to perform contour deduction, or modeling the shape and motion state of the target contour through a state space model.
[0044] S3: When the height of the target power tower in the key frame image exceeds a certain ratio, the optimal ratio is set to three quarters. The key parts identification method of the power tower is adopted, that is, the key points on the target power tower are identified through a convolutional neural network to determine the angle and distance of the drone relative to the target power tower. This method transfers target detection from the power tower to the key points on the target power tower, from single target identification to multi-target identification, and segments the key points in the target power tower. Through multiple segmentation positions, the angle and distance of the drone relative to the target power tower are determined, and the binocular ranging function on the drone is enabled to assist the drone flight. Key points can include: insulators, cable hanging points, tower spires, tower wings, etc.
[0045] Specifically, when the UAV is close to the target power pole tower, that is, when the target power pole tower selection box height is close to the image height, it is judged that the distance between the UAV and the target power pole tower is close. To prevent the risk of collision, the speed of the UAV is reduced, and a close-range and fine-precision method for identifying key parts of the power pole tower is used to approach the target power pole tower.
[0046] In terms of identifying key parts of power poles and towers, the focal length of the camera in the gimbal device is first adjusted. The focal length is first reduced to allow the drone to get closer to the target power pole. After approaching to a certain distance, the focal length is increased. This allows the drone to better detect the structure of the entire target power pole, while also reducing the interference of irrelevant objects around the target power pole on target detection.
[0047] The binocular ranging function is mainly to achieve the obstacle avoidance function. The binocular camera uses the visual odometry (VO) technology. Through the images obtained by the left and right binoculars, the distance between the key points of the target power poles around the drone and the drone is inferred. Compared with the traditional optical flow + ultrasonic technology, the visual odometry can also build a three-dimensional model around the drone, and through continuous images, it can judge the relative movement between itself and the environment and estimate its own movement status. Accurately measure the relative position between itself and the target power pole, so as to achieve the function of precise obstacle avoidance and ensure the safety of the drone during cruising. At the same time, due to different safe working distances for different high-voltage lines, the safe distance of the obstacle avoidance function needs to be set according to the actual working environment.
[0048] S4: According to the distance and angle data between the drone and the target power tower obtained in step S3, the drone moves to a direction perpendicular to the transmission line connected to the power tower, and then makes the drone higher than the target power tower and shoots key points on one side of the target power tower. After shooting one side of the target power tower, the drone adjusts its position to shoot key points on the other side of the target power tower. After that, the drone selects power towers for inspection in sequence according to the numbers.
[0049] S5: Upload the captured image obtained in step S4 to the image library, and add feature information to the captured image, the feature information includes but is not limited to the shooting time, the number of the captured power pole tower, the key point number, the selection frame coordinates and other feature information, to provide information for subsequent defect troubleshooting and other needs.
[0050] like Figure 2 The flowchart of a method for recognizing power pole tower images by a drone provided in an embodiment of the present invention. The convolutional neural network recognition of power pole towers in step S2 includes the following steps:
[0051] S100: Obtain a first image and pre-process it;
[0052] S200: performing image segmentation on the preprocessed first image using a convolutional neural network, and selecting features of the power pole using a bounding box, wherein parameters of the bounding box include a probability and a confidence level of the power pole in the bounding box;
[0053] S300: Use a convolutional neural network to output the target detection result, select the most optimal frame through non-maximum suppression, and obtain the location of the power tower.
[0054] Specifically, if Figure 3 A method for unmanned aerial vehicle to recognize and frame power pole tower images includes the following steps:
[0055] S201: performing grid segmentation on the preprocessed first image through a convolutional neural network, where the number of grids is set according to the distance between the drone and the power pole tower;
[0056] S202: distinguishing the shape of the power tower and the surrounding environment, marking the selection frame coordinate information of the power tower, and constructing a power tower image database;
[0057] S203: using the constructed power tower image database to train the convolutional neural network, including the convolutional layer, BN layer, LR layer, pooling layer, 1*1, 7*7 and other convolution kernels of the data, to obtain the key weight information of the convolutional neural network;
[0058] S204: Calibrate the pixel coordinates of the power tower in the first image using the trained convolutional neural network.
[0059] like Figure 4 A schematic diagram of a neural network structure for drone detection and autonomous cruising provided in an embodiment of the present application.
[0060] The convolutional neural network includes a first convolutional layer, a feature extraction layer, and multiple convolutional layers connected in sequence. The feature extraction layer includes a VBL layer, a Resn layer, and a pooling layer. The VBL layer includes a VGG convolutional layer, a Batch Normalization, and a Leaky relu layer.
[0061] During the long-distance autonomous cruising of the drone, the take-off point of the drone is set to be far away from the power pole. This is a single target detection task, and there is no need to consider the problem of overlapping and resolving multiple targets. At the same time, in order to make the drone approach the power pole relatively quickly, a faster algorithm is required. Therefore, a fully convolutional neural network is used to identify the target power pole, and the front of the drone is filmed through video.
[0062] The fine-precision recognition process requires a high degree of detail in the power tower structure and key points, so a higher resolution is used to detect the target. At the same time, due to the low navigation speed of the drone during fine-precision recognition, a higher amount of computational recognition work is allowed. Therefore, the number of convolutions can be appropriately increased, or fully connected layers and pooling layers can be added to achieve the purpose of accurate recognition.
[0063] Specifically, the specific process of power pole tower recognition through convolutional neural network is as follows:
[0064] The preprocessed first image is input into the first convolution layer to obtain a first feature map, and its function is mainly to input the first image to be detected whose resolution meets the preset resolution into the first convolution layer to obtain the first feature map.
[0065] Feature extraction layers such as Conv_layers 1 are composed of VBL layer, Resn layer, and pooling layer, among which the VBL layer is mainly composed of VGG convolution, BatchNormalization and Leaky relu layer.
[0066] The Resn layer is mainly the residual structure of the ResNET network, where n represents the number of layers.
[0067] Connect multiple convolutions after the feature extraction layer, such as Figure 4 CONV_layers2, CONV_layers3, etc. in , in order to generate feature maps of different scales. Then, bounding boxes of different scales and sizes are constructed on these feature maps, and detection and classification are performed respectively to generate multiple bounding boxes that preliminarily meet the conditions. The bounding box contains location attributes, as well as the probability and confidence of correct target recognition. Since only a single target is recognized when identifying power poles and towers, the target possibility and confidence parameter dimensions contained in the bounding box are set to 1.
[0068] Specifically, the position attribute of the bounding box is represented by four values (x, y, w, h), where (x, y) represents the offset of the center position of the bounding box relative to the coordinates of the upper left corner of the grid. The calculation process is as follows:
[0069]
[0070] Where xc, yc represent the coordinates of the center position of the bounding box, wi, hi represent the original height of the screen, Xcol, Yrow represent the coordinates of the upper left corner of the grid where the center of the bounding box is located, and its value range is (0, n), where n represents the number of grids.
[0071] (w,h) represents the proportion of the bounding box to the entire image, and the calculation formula is as follows:
[0072]
[0073] The above coordinates all have a value range of [0,1]. The purpose of this is that power tower target detection is a classification problem, and using variables between [0,1] is more suitable for the classifier we use.
[0074] The video is divided into frames, and key frame images are extracted from the video. The key frame images are converted into grayscale images to reduce the image data dimension, and then edge points are extracted to retain key pixels to minimize the amount of calculation.
[0075] Since only a single target is detected, the softmax function is no longer used as the classifier. Instead, the binary classification function logistic regression is used as the classifier.
[0076] After that, the bounding box of the output result of the convolutional neural network model is processed by non-maximum suppression. This process is an iterative-traversal-elimination process. Specifically, the scores of all bounding boxes are sorted, and the highest score and its corresponding bounding box are selected. Traverse the remaining bounding boxes, and if the overlapping area with the current highest score bounding box is greater than a certain threshold, we will delete the bounding box. Select the highest score from the unprocessed bounding box and repeat the above process. Finally, use NMS (non-maximum suppression) to transmit the coordinates of the optimal bounding box to the worker in a visual form, and record the bounding box data to help the drone route planning.
[0077] The training process of a convolutional neural network consists of the following steps:
[0078] S11: Using on-site shooting, shooting the power poles and towers according to different lighting conditions, weather conditions, distances and angles, and calibrating and classifying the power poles and towers images according to the on-site weather and lighting conditions to build a power poles and towers image library;
[0079] S12: Improve the performance of the algorithm by using a variety of data augmentation methods, including horizontal flipping, cropping, zooming in and out, and inputting image data of different sizes and shapes;
[0080] S13: manually mark the power pole tower area, use the labeling tool labeimg to mark the area of interest, and use the CVAT calibration tool to mark the key points on the power pole tower;
[0081] S14: Train the convolutional layer of the convolutional neural network on the constructed power tower image library dataset.
[0082] The basic principles and main features of the present invention and the advantages of the present invention are shown and described above. For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention; therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is limited by the attached claims rather than the above description. Therefore, it is intended to include all changes within the meaning and scope of the equivalent elements of the claims in the present invention, and any figure marks in the claims should not be regarded as limiting the claims involved.
[0083] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.
Claims
1. A method for detecting and autonomously cruising a power tower using an unmanned aerial vehicle based on image recognition, characterized in that: The following steps are involved: S1: manually setting the navigation direction of the drone equipped with the gimbal device, taking a picture of the power transmission line in front of the drone to obtain a first image, extracting the features of the power tower using the constructed convolutional neural network after preprocessing the first image, identifying the positions of all the power towers in the first image and numbering them in sequence; S2: The drone flies toward the selected target power pole and records a video, extracts key frame images from the recorded video at a certain time interval, recognizes the key frame images through the convolutional neural network, selects and marks the identified target power pole, and compares them to determine the distance between the drone and the target power pole; S3: When the height of the target power pole tower in the key frame image exceeds a certain ratio, the key points on the target power pole tower are identified by the convolutional neural network to determine the angle and distance of the drone relative to the target power pole tower; S4: According to the distance and angle data between the drone and the target power tower obtained in step S3, after the drone moves to a direction perpendicular to the transmission line connected to the power tower, the drone is made higher than the target power tower and photographs key points on one side of the target power tower. After photographing one side of the target power tower, the drone adjusts its position to photograph key points on the other side of the target power tower. After that, the drone selects power towers for inspection in sequence according to the numbers; S5: Upload the captured image obtained in step S4 to the image library, and add feature information to the captured image; The convolutional neural network in step S1 for identifying power poles and towers includes the following steps: S100: Obtain a first image and pre-process it; S200: the convolutional neural network performs image segmentation on the preprocessed first image, and uses a bounding box to select features of the power pole tower, wherein the parameters contained in the bounding box include a probability and a confidence level of the power pole tower in the bounding box; S300: using the convolutional neural network to output the target detection result, filtering out the most optimal frame through non-maximum suppression, and obtaining the power tower position; In step S3, when the height of the target power tower in the key frame image exceeds three quarters, the key points on the target power tower are identified by the convolutional neural network to determine the angle and distance of the drone relative to the target power tower.
2. The method for detecting and autonomously cruising a power tower using an unmanned aerial vehicle based on image recognition according to claim 1, characterized in that: The frame selection of the power pole tower by the convolutional neural network includes the following steps: S201: Segmenting the preprocessed first image into grids by using the convolutional neural network, wherein the number of grids is set according to the distance between the drone and the power tower; S202: distinguishing the shape of the power tower and the surrounding environment, marking the selection frame coordinate information of the power tower, and constructing a power tower image database; S203: training the convolutional neural network using the constructed power pole tower image database to obtain key weight information of the convolutional neural network; S204: Calibrate the pixel coordinates of the power pole tower in the first image and select the power pole tower by using the trained convolutional neural network.
3. The method for detecting and autonomously cruising a power tower using an unmanned aerial vehicle based on image recognition according to claim 1, characterized in that: The convolutional neural network includes a first convolutional layer, a feature extraction layer, and multiple convolutional layers connected in sequence, the feature extraction layer includes a VBL layer, a Resn layer, and a pooling layer, and the VBL layer includes a VGG convolutional layer, a BatchNormalization layer, and a Leakyrelu layer.
4. The method for detecting and autonomously cruising a power tower using an unmanned aerial vehicle based on image recognition according to claim 1, characterized in that: The pre-training process of the convolutional neural network includes the following steps: S11: Using on-site shooting, shooting the power poles and towers according to different lighting conditions, weather conditions, distances and angles, and calibrating and classifying the power poles and towers images according to the on-site weather and lighting conditions to build a power poles and towers image library; S12: Improve the performance of the algorithm by using a variety of data augmentation methods, including horizontal flipping, cropping, zooming in and out, and inputting image data of different sizes and shapes; S13: manually mark the power pole tower area, use the labeling tool labeimg to mark the area of interest, and use the CVAT calibration tool to mark the key points on the power pole tower; S14: Train the convolutional layer of the convolutional neural network on the constructed power tower image library dataset.
5. The method for detecting and autonomously cruising a power tower using an unmanned aerial vehicle based on image recognition according to claim 1, characterized in that: In step S3, when the drone determines the angle of the target power pole, the binocular ranging function carried by the drone is enabled to assist the drone flight.
6. The method for detecting and autonomously cruising a power tower using an unmanned aerial vehicle based on image recognition according to claim 1, characterized in that: The distance between the UAV and the target power pole tower is determined by using a gradient search to perform a contour deduction method or a state space modeling method to model the shape and motion state of the target contour.
