Target Tracking Based Upon Dominant Orientation Template and Kalman Filter

Abstract

Due to the proliferation of surveillance systems in defense as well as civilian sectors, target tracking field has always remained adorable to researchers of the computer vision community. Template matching based approaches are a conventional way to solve target tracking problems. Ironically such approaches are computationally complex due to a requirement of repetitive arithmetic operations. A number of attempts have been made to reduce this computational overhead; Dominant Orientation Template (DOT) based template matching has given a new dimension to solve problems of such a kind using logical operations only. It has been observed that the performance of template matching based approaches degrades drastically in the presence of occlusion. DOT-based template matching is also not an exception to this. Proposed approach introduces a novel target tracking framework having severe occlusion handling capacity by integrating DOT-based template matching and Kalman Filtering. In the present approach, a small target search window is formulated around the Kalman estimated location of the considered frame; rather than exploring the whole frame. DOT-based template matching works inside this window only. A parameter known as Kalman error is defined here which is considered as the measure of occlusion. Initially, DOT calculated location is assumed as measured value but in case of occlusion when this Kalman error becomes sufficiently large the Kalman estimated location is regarded as the measurement. A customized dataset having occluded targets in FLIR as well as visible video sequences is generated to provide a robust testbed for the proposed approach.