Method for Detecting and Tracking Foreign Objects in Substation Videos Based on Embedded AI

Foreign object monitoring is very important to maintain the safe running of substations. The development of intelligent video monitoring technology provides a solution for real-time warning and foreign object tracking in substations. Intelligent video recognition technology is gradually replacing manual monitoring. However, the current video recognition algorithms, such as moving object detection, deep learning object detection and tracking, have their own advantages and disadvantages, which cannot meet the targeted recognition of foreign objects of interest in substations or it is difficult to achieve real-time detection on embedded devices. Moving object detection is fast. YOLOv4 can locate and recognize the objects of interest with high recognition rate. And tracking algorithm has fast speed as well as good tracking continuity. Combining these three algorithms, this paper proposes a method based on YOLOv4 for foreign object monitoring and tracking in substation videos. And experiments show the method performs well in the detection and tracing of ILSVRC VID dataset and actual substation videos. It can realize front-end detection and real-time monitoring on NVIDIA Jetson Xavier NX embedded development board.