Computer Vision-Based Pointer-Style Mechanical Dial Reading Method for LNG Terminal Safety

As the demand for liquefied natural gas (LNG) continues to grow, the number of LNG terminals is increasing. To ensure safe operation at LNG terminals, various types of metering instruments with pointer-style dials are installed to monitor the real-time operating status of the LNG facilities. aiming at the problems of low accuracy, low real-time detection efficiency and high risk of traditional manual reading methods, a deep learning pointer instrument automatic reading method based on PP-PicoDet and BiSeNetV2 is constructed. Firstly, the PP-PicoDet object detection algorithm is used to locate the position of the instrument dial in the complex background image. Then, the BiSeNetV2 semantic segmentation algorithm is used to segment the pointer and scale areas from the background image, to fit a circular dial using the least-squares-circle function and fit the pointer as a straight line using the curve-fit function, Finally, the angle method algorithm developed independently is used to calculate the instrument reading. The algorithm is experimentally verified by selecting the liquid level meter and pressure gauge. The results show that the reading error is about 2%, and the average recognition time is about 415ms, which has good accuracy and speed, and meets the actual requirements of instrument reading recognition, realizing the intelligent real-time safety monitoring of LNG terminals.