Object Tracking Using Appearance Model Based on Feature Extracted from Multi-Scale Image Feature Space with Data-Independent Basis

Abstract

It's an arduous task to write an effcx: tivc and e ffic ient a lgorithm for robust object tracking. Object tracking depends on factors like variation in pose, motion blurring, change in illumination and occlusion. Many objects tracking a lgori thms monitoring samples in previous frames and update the models. While accomplishment has been achieved, still many issues remain needed to be entertained. Firstly, adaptive appearance algori thms are dependent on data, and a heavy amount of informmion is needed to learn for online algorithms at the outset. Secondly, drift problem is experienced by online tracking algorithms while detecting the o bject in fa st motion. Miss-a ligned samples are added which reduce e rlicicncy and accuracy o f appearance model caused by selftaught learning. In this research. I propose an elementary robust object tracking algorithm by using an appearance model. This model based on features which can be obtained by an image feature space on the basis of data -independent, whjch is more effect ive and efficient. This model uses no adapt ive random projections, which usually retain the stnJcture of an object. To extract features accurately and efficiently for an appea rance model, sparse measurement matrix is used for it. TIle foreground and background samples of targel object compress with the help o f sparse measurement matrix. The target objcct is tracked by Nai've Bayes classi fier for binary classification by compressing. The proposed o bject tracking algorithm is compress ive and perfonns favorably well 1Il a mean of efficiency. accuracy. and robustness.