Dynamic Tracking and Reading Algorithm of Intelligent Patrol Instrument based on YOLOv5 Neural Network

The digital reading of instrument by inspection robot is the current trend of development; the existing research mainly focuses on the dials reading of the collected instrument images, and the lack of intelligent acquisition process of instrument images, resulting in complex image acquisition process, low image definition and poor image quality. Thus to improve the effective coordination of acquisition equipment in the inspection operation and conduct the research of relevant algorithms has become the current focus, especially in the petrochemical scenario, the inspection operations require a more flexible inspection method. This paper proposes a tracing method for instrument detection which based on YOLOv5 neural network and deviation angle compensation. The target position information is obtained by the method of target detection, and the deviation angle is calculated. The target tracking algorithm in the tracking module is studied so that the dial is placed in the center of the overall image with an appropriate proportion. It has been verified that YOLOv5 neural network can better complete the instrument positioning work, and the obtained deviation error of the target position is less than 5% of the screen proportion. It can effectively complete the instrument target tracking work and obtain the ideal dial to be identified. This research has certain engineering application value for improving the working reliability of inspection robots, reducing the maintenance cost in the later stage, and promoting the unmanned inspection.