Method for carrying out angle estimation on tower lead

Abstract

According to the technical scheme, the method is characterized by comprising the following steps that target detection is carried out through a Yolo target detection model, and target detection and recognition are carried out on a tower head, a tower body, a cross arm, an insulator, parallel wires, vertical wires, a wire with the positive slope and a wire with the negative slope of an electric power tower; after the leads of various postures are recognized, a square frame is drawn for the leads through target detection, the square frame is close to the leads as much as possible, and then the pointing direction and angle of the leads are obtained. According to the invention, dual models are used for combined judgment on the recognition of the lead, and the accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent recognition, and particularly relates to a method for estimating the angle of tower conductors. Background Art

[0002] When an unmanned aerial vehicle (UAV) conducts inspection over power transmission towers, especially when the accurate GPS positions of all towers on the line are not provided, the inspection is prone to low efficiency due to the lack of guidance.

[0003] Generally speaking, the following problems exist currently: If an image segmentation algorithm is adopted, it is too heavy and consumes too much computing power to obtain real-time performance. If lidar and multi-sensor fusion are adopted, for some inspection UAVs, through the fusion of lidar with multiple sensors such as cameras and infrared, three-dimensional space detection and conductor recognition are carried out. Multi-sensor fusion can improve the recognition effect, but it has a high cost, and the sensor fusion algorithm is complex, increasing the difficulty of data processing and real-time performance. If SLAM and path planning are adopted, in the UAV conductor inspection, combined with SLAM technology, it is used for the path planning and positioning of the UAV. This technology is helpful for the real-time detection and recognition of the conductor angle direction, but there are still deficiencies in real-time performance and algorithm complexity.

[0004] Therefore, the current defects can be summarized as follows: The real-time processing is difficult. The processing resources of the UAV are limited, and conductor detection and angle estimation usually require real-time calculation. Especially during high-speed flight, higher requirements are imposed on the detection accuracy and response speed. The existing image segmentation models are large and time-consuming, and it is difficult to meet the real-time requirements. If multi-sensor fusion is adopted, data synchronization and fusion calculation will further increase the difficulty of real-time processing.

[0005] The cost is high and the equipment requirements are high. The cost of the UAV inspection system combining lidar and other multi-sensors is high, and ordinary inspection UAVs do not have the conditions to use these expensive sensors. And using high-precision vision equipment will also significantly increase the inspection cost, affecting the popularization and application. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for estimating the angle of tower conductors to solve the problems raised in the above background art.

[0007] To achieve the above purpose, the present invention provides the following technical solution: A method for estimating the angle of tower conductors, comprising the following steps: performing target detection through a Yolo target detection model to detect and identify the tower head, tower body, cross arm, insulator, parallel conductors, vertical conductors, conductors with a positive slope, and conductors with a negative slope of the power transmission tower; after identifying conductors in various postures, drawing a box for the conductor by the target detection and making the box as close to the conductor as possible, and then obtaining the direction and angle pointed by the conductor.

[0008] Preferably, the Yolo object detection model includes an object detection module and an object recognition module, and the object detection module and the object recognition module respectively detect and recognize the tower head, tower body, cross arm, insulator, parallel conductors, vertical conductors, conductors with a positive slope, and conductors with a negative slope of the power pole tower.

[0009] Preferably, the object detection module and the object recognition module are respectively connected to a server, and the server is connected to a data storage module, and the data storage module is used to store the data detected and recognized by the object detection module and the object recognition module.

[0010] Preferably, the server is connected to the data storage module through a communication module, and the communication module includes a WIFI sub-module, a 4G sub-module, and a 5G sub-module.

[0011] Preferably, the data storage module is connected to a data encryption module, and the data encryption module is connected to a data decryption module.

[0012] Preferably, the object detection module and the object recognition module respectively include a bounding box coordinate management sub-module, a confidence level and class ID management sub-module, a class label mapping sub-module, and an image data management sub-module.

[0013] Preferably, the image data management sub-module includes an image segmentation unit, and the fields included in the image segmentation unit are instance segmentation mask data, bounding box information, and class label mapping; the mask data of the image segmentation unit is a binary image, and the binary image is adjusted to the same size as the original image and superimposed with the original image for visualization.

[0014] Preferably, the object recognition module is connected to a pose estimation module, the pose estimation module processes key points and skeleton structures, and the pose estimation module is connected to a box correction module, and the box correction module is used to correct the drawn box.

[0015] Compared with the prior art, the beneficial effects of the present invention are:

[0016] In the identification of wires, the present invention uses a combined judgment of dual models to improve the accuracy. When the drone conducts inspections over the power transmission towers, it uses an algorithm for automatically identifying directions and flying correctly towards the next-level tower. With the present invention, the drone can be guided to fly autonomously towards the next-level tower, and then repeat the process to inspect a sufficiently long line before the battery runs out. In addition, the real-time processing of the present invention is less difficult and requires low costs, overcoming the defect of increasing inspection costs by using high-precision vision devices. The present invention conducts target detection through the much simpler yolo model in terms of computing power, and conducts target detection and identification on the tower head, tower body, cross arm, insulator, parallel wires, vertical wires, wires with a positive slope, and wires with a negative slope of the power transmission tower. After identifying wires in various postures, the target detection will draw a box around the wire, and this box will be as close as possible to the wire, so that the direction pointed by the wire can be quickly known. Detailed implementation manners

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0018] Embodiment 1

[0019] The present invention provides a technical solution: a method for estimating the angle of tower wires. A method for estimating the angle of tower wires includes the following steps: conducting target detection through the Yolo target detection model, and conducting target detection and identification on the tower head, tower body, cross arm, insulator, parallel wires, vertical wires, wires with a positive slope, and wires with a negative slope of the power transmission tower; after identifying wires in various postures, the target detection draws a box for the wire and makes the box as close as possible to the wire, and then obtains the direction and angle pointed by the wire.

[0020] Among them, the Yolo target detection model includes a target detection module and a target identification module. The target detection module and the target identification module respectively conduct detection and identification on the tower head, tower body, cross arm, insulator, parallel wires, vertical wires, wires with a positive slope, and wires with a negative slope of the power transmission tower.

[0021] Among them, the target detection module and the target identification module are respectively connected to a server, and the server is connected to a data storage module. The data storage module is used to store the data detected and identified by the target detection module and the target identification module.

[0022] Among them, the server is connected to the data storage module through a communication module. The communication module includes a WIFI sub-module, a 4G sub-module, and a 5G sub-module.

[0023] Among them, the data storage module is connected to a data encryption module, and the data encryption module is connected to a data decryption module.

[0024] Among them, the target detection module and the target recognition module respectively include a bounding box coordinate management sub-module, a confidence level and class ID management sub-module, a class label mapping sub-module, and an image data management sub-module.

[0025] Among them, the image data management sub-module includes an image segmentation unit. The fields included in the image segmentation unit are instance segmentation mask data, bounding box information, and class label mapping. The mask data of the image segmentation unit is a binary image, which is adjusted to the same size as the original image and overlaid with the original image for visualization.

[0026] Among them, the target recognition module is connected to a pose estimation module. The pose estimation module processes key points and skeleton structures. The pose estimation module is connected to a box correction module, and the box correction module is used to correct the drawn square box.

[0027] Among them, in the identification of wires in the present invention, a dual model combination judgment is used to improve the accuracy. When the drone conducts inspections over power transmission towers, an algorithm for automatically identifying the direction and being able to fly correctly towards the next-level tower is used. The present invention can guide the drone to fly autonomously towards the next-level tower, and then repeat the process. Before the battery runs out, it can inspect a sufficiently long line. In addition, the real-time processing of the present invention has a low difficulty and requires a low cost, overcoming the defect of increasing the inspection cost by using high-precision vision devices. The present invention performs target detection on the tower head, tower body, cross arm, insulator, parallel wires, vertical wires, wires with a positive slope, and wires with a negative slope of the power transmission tower through the much simpler yolo model in terms of computing power. Then, after identifying wires in various postures, the target detection will draw a square box for the wire, and this square box will be as close as possible to the wire, so that the direction pointed by the wire can be quickly known.

[0028] Among them, the present invention has simplicity: less code volume. The present invention has high performance: compared with the image segmentation algorithm (Segmentation), only two Yolo object detection models are used. On an edge device, the former takes 10 - 30 seconds for one detection, while the latter only takes 300 - 500 ms. The present invention has accuracy: through field verification, the accuracy and precision on ordinary poles and corner towers are high enough to meet the requirements of controlling the drone to fly in a specified direction. The present invention independently researches and develops a wire model: for the recognition of wires, a dual model combination judgment is used to improve the accuracy. The two models are also trained based on the independently developed algorithm and the proprietary data, which are unique. The present invention can achieve fast and accurate: the combination of the efficiency, performance, and accuracy of this algorithm is relatively advanced and has been commercially verified in the customer's field.

[0029] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for estimating the angle of a tower conductor, characterized in that: The following steps are involved: The Yolo target detection model is used to detect and identify the tower head, tower body, cross arm, insulator, parallel wires, vertical wires, wires with positive slopes, and wires with negative slopes of power towers. After identifying wires in various postures, the target detector draws a box for the wire and makes the box as close to the wire as possible, and then obtains the direction and angle of the wire.

2. The method for estimating the angle of a tower conductor according to claim 1, wherein: The Yolo target detection model includes a target detection module and a target recognition module, which respectively detect and recognize the tower head, tower body, cross arm, insulator, parallel wires, vertical wires, wires with positive slopes, and wires with negative slopes of the power tower.

3. The method for estimating the angle of a tower conductor according to claim 2, wherein: The target detection module and the target recognition module are respectively connected to a server, and the server is connected to a data storage module, and the data storage module is used to store data detected and recognized by the target detection module and the target recognition module.

4. The method for estimating the angle of a tower conductor according to claim 3, wherein: The server is connected to the data storage module via a communication module, and the communication module includes a WIFI submodule, a 4G submodule, and a 5G submodule.

5. The method for estimating the angle of a tower conductor according to claim 3, wherein: The data storage module is connected to a data encryption module, and the data encryption module is connected to a data decryption module.

6. The method for estimating the angle of a tower conductor according to claim 2, characterized in that: The target detection module and the target recognition module respectively include a bounding box coordinate management submodule, a confidence and category ID management submodule, a category label mapping submodule and an image data management submodule.

7. The method for estimating the angle of a tower conductor according to claim 1, characterized in that: The image data management submodule includes an image segmentation unit, and the fields included in the image segmentation unit include instance segmentation mask data, bounding box information, and category label mapping; the mask data of the image segmentation unit is a binary image, and the binary image is adjusted to the same size as the original image and superimposed with the original image for visualization.

8. The method for estimating the angle of a tower conductor according to claim 2, characterized in that: The target recognition module is connected to a posture estimation module, which processes key points and skeleton structures. The posture estimation module is connected to a frame correction module, which is used to correct the drawn frame.