Development of Object Tracking System Utilizing Camera Movement and Deep Neural Network

Object tracking is an essential field of study in the field of computer science with its applications ranging from video surveillance to autonomous driving. Object tracking also has been a challenging task in the field of computer vision, the main difficulties stemming from issues in the image such as background distractions, multiple tracking scales, occlusion etc. There are numerous approaches to keep track of objects moving in the camera frame. The new objects appearing in the camera frame are registered for tracking and the ones that are lost or disappeared from the frame are deregistered. This method does not certainly help if the user wants to track one specific object throughout, even if the object is moving out of the camera frame. This paper presents a novel approach to track a specific user defined object, utilizing camera movement and custom-trained deep neural network so that even if the object is about to move out of the camera frame, the camera orientation is changed in order to keep that object well centered in the frame. In this approach the camera is mounted on a 2-axis motorized gimbal and video stream is fed to a computer which runs YOLOv4, a one-stage object detection model. The user defined object, if detected, its position relative to overall frame is calculated and necessary change to the camera orientation is made to keep the object at the center of frame. The proposed object tracking algorithm works flawlessly as long as object detection algorithm performs accurately.