Ohta based covariance technique for tracking object in video scene

In this paper we propose Ohta based covariance method for tracking an object in various challenging situations. In the proposed method covariance matrix is used as the region descriptor. In addition to this other elements of feature vector are colour moments, derivatives of the region and the pixel position. We incorporated a model update algorithm based on Riemannian geometry for updating covariance matrix. This method has been applied to four different conditions and the resulting experimental results show the robustness of the technique against occlusion, camera motion appearance and illumination change. Also the performance of this technique is compared with other existing techniques such as the covariance method with RGB features and histograms based method in terms of detection rate and show the superiority against other two.