Automatic Object Tracking using Deep Learning Technique

Abstract

Due to the rapid increase of necessity in security and military applications, surveillance systems have
become a necessary area of study. Asking human operators to keep watch for long hours is not only a
cumbersome task but it also increases the chance of error. Thus, to assist human operators identify events
which are important, Automatic Object Tracking is proposed. An object is tracked by, firstly, detecting
the object using any of the various object detection methods in frames present in the input video. These
methods make use of the spatial domain, temporal changes, presence etc. of the objects present. Every
object is then tracked using any of the various methods. This can be used for monitoring traffic,
animation, robot vision and video surveillance. In the proposed system, YOLO v2 is being used for Object
Detection and Kalman Filter along with Non-Maximum Suppression will be used for Automatic Object
Tracking.