Adaptive Algorithm For Fast And Accurate Video Object Tracking Using SIFT And BMA For Slow And Rapid Movements

A novel adaptive video object tracking algorithm is proposed using intensity based and feature based techniques which results in improved reliability and at the same time it aims to reduce computational complexity. This is achieved by intelligently combining SIFT and BMA and making it adaptive to the speed of the moving object. Tracking frame by frame using SIFT which is successful for detecting moving objects, especially along the camera optical axis where the object scale changes; and even in case of partial affine distortion and partial occlusion, but SIFT feature extraction and descriptor matching is a very computationally intense process. In this work SIFT is applied on the first frame and then on every nfx th frame. The computationally simple BMA is applied in the intermediate frames for object tracking. The value of nfx is variable and is determined from the object speed factor ‘s’ by means of the proposed graph. The work presented also gives the optimized value of ‘n’ for the application of SIFT with respect to the speed factor ‘s’ of the moving object. KeywordsBlock Matching, SIFT, spatial transform, Object Tracking, Feature Matching.