Thermal Infrared Tracking System with YOLOv4 Algorithm and Visual Servoing Method

Abstract

This paper presents design and implementation of human object tracking system using an infrared thermal camera and servo-motor-controlled pan-tilt system on a single board computer. The implementation of the system is aimed to develop air to ground EOTS (Electro Optical Targeting System) to track moving object in a low light condition. A thermal camera is used to obtain the visual information from an object in a low light condition. By utilizing visual servoing method, YOLOv4 algorithm is used to give information about object’s position on screen which will be used to control the pan-tilt servo motor to track the object movement. The detection process is performed from approximately 5 meters from the ground. This is to simulate the air to ground environment. This system consists of the acquisition subsystem, data processing subsystem, user interface subsystem, and actuator subsystem. This paper will discuss the process of design and implementation of each subsystem, system integration and the system testing. The experiment result shows that the system is able to track an object with a 14 Frame Per Second (FPS) framerate and human objects could be detected with an 96,61% average precision score.