Robust Visual Object Tracking

Abstract

It is very easy for humans to recognize the object and to follow it till their sight range, but with the advancement of technology, we want to take this work through machines so that we can get better results according to our desire. So, with this need, the computer vision field came out of the box by using its sub-fields like object detection and visual object tracking. Tremendous efforts are being done by researchers in a field of object tracking, but it is still open to be explored because the challenges of visual object tracking still exist and this thesis also deals with the visual object tracking challenges. We consider the main challenge of tracking by our motivation i.e., Occlusion and illumination variation. So, in this regard, we select the state of the art algorithm name with “Adaptive Correlation Filters with Long-Term and Short-Term Memory for Object Tracking” in which three correlation filters were proposed 1) LongTerm Filter 2) Translation Filter 3) Scale filter. These filters work outstanding in most of the video sequence, but we have found that their performance degrades for some of the video sequences bearing challenges of occlusion and illumination variation. In order to solve these problems, we incorporate one more filter that is Kalman filter to the algorithm; enhanced algorithm yields better results as compared to its counterpart method when video sequences having challenges of occlusion and illumination variation is given to the proposed method has been tested on standard dataset i.e., Object Tracking Benchmark 13 containing 49 video sequences with different challenges the comparison of the proposed method with its base method i.e., existing selected method. The proposed method highlights its effectiveness both quantitatively and qualitatively, especially in occluded and varying illumination environment.