Real Time Dynamic Object Tracking with a Pan and Tilt
Camera Setup
Harsha De Silva* 1 , Ruchira Kalhara Mahaliyana 1 , Isuru Nawinne 1 and Hiranya Jayakody 2
1
Department of Computer Engineering, Faculty of Engineering, University of Peradeniya, Sri Lanka
2
School of mechanical and manufacturing engineering, faculty of Engineering, University of New South Wales.
*E-mail: [email protected]
Abstract: This paper presents the implementation and development of an action tracking camera. The final
objective is to develop a pan and tilt camera platform which can control itself to follow and focus on a moving
object automatically the same way any living creature watches a moving object through the naked eye. In this
research, a pre-defined object is detected and localized from a video stream captured by a camera. Based on the
relative position of the object in the image frame, pan and tilt commands are generated by a novel algorithm to
track the object. A deep learning approach is used to detect and localize the object from the video stream. Based
on its output, the camera is controlled so the object is brought closer to the center of the image frame as much
as possible. Using the YOLO-v3 object detection algorithm, a football was tracked with a frame rate of 30fps on
Tesla P4 GPU with an average precision of 54%. The camera controlling algorithm works at a frame rate of 7fps
with camera movements.
Keywords - real time tracking , pan & tilt camera, moving object tracking, image processing, convolutional
neural network,deep learning
22