A UAV-Based Machine Vision Algorithm for Industrial Gauge Detecting and Display Reading

Chun Li, Dehua Zheng, Lizheng Liu, Xiaochen Zheng

Abstract: Unmanned Aerial Vehicle (UAV) has overwhelming superiority on the completion of difficult missions in the industrial production or implementation scenarios. Its brilliant navigation and on-board perception abilities endow the aerial platform a considerable potential in the industrial applications. Since manipulation stability and computational capacity of UAVs are crucial for industrial missions, it is challenging to develop a reliable and safe UAV platform for indoor industrial operation. Focusing on the measurement of industrial-standard gauges, we propose a vision algorithm which is capable of fast detecting the industrial-standard gauges and the readings and integrate the algorithm into a quadrotor drone platform with indoor and outdoor navigation. In our work, we demonstrate how to improve the simplicity and efficiency of the UAV-based visual recognition by implementing and adjusting a YOLO v3 framework with Darknet [1]. Moreover, our vision algorithm is combined with the image geometric correction module and the gauge detecting and reading module to overcome the detection problems caused by the harsh industrial conditions, such as an obscure image and the under-exposure condition. And the results show that accuracy of detection in the experimentation is sufficient for industrial missions.