Design of Target Tracking Ball Gear Pan-Tilt System Based on Machine Vision

The trend of intelligent and unmanned industries benefited from the rapid development of information technology is more and more obvious. The unmanned combat platform has become a new combat force in anti-terrorism operations, relying on sensor, machine vision, communication, information fusion and other technologies. A design scheme of pan-tilt target tracking system based on machine vision is proposed here for the visual perception requirements of unmanned combat platform. A combination of YOLOv4-tiny and DeepSORT is used to identify and track personnel with cameras mounted on ball gears. When the target moves to the edge of view, the position information will be sent to the control circuit through serial communication, and the camera will be adjusted with the PID control algorithm to keep the target in the center of view. The experimental results showed that this system can achieve target tracking effectively in real time and the tracking success rate can reach 80% by optimizing with K-Means++ algorithm and PID controller.