Region based multiple object tracking with LSTM supported trajectories

Abstract

Object Tracking is the growing field in the area of computer vision with its demands in various areas in monitoring and surveillance. Areas of surveillance can be improved with proper and efficient trackers which can ensure the safety of people on roads, the safety of children in school, and in many other areas. Object tracking is a super-set of elements object classification, object detection, etc. In this context, literature is being reviewed for object tracking using deep learning architectures and standard object detection algorithms. Most of the work done consists of tracking based on visual features so we have worked on region-based features. A method is proposed for tracking only on the basis of region features extracted with the help of intersection over union and prediction of trajectories with the help of lstm in case of occlusion. Comparison for the same is being carried out with the traditional centroid tracking algorithm