Tracking people through selected blocks using correlation and optimized similarity measure OSSD

Abstract

This letter correspondence exposes results of a simple and appropriate technique for tracking a moving person in a stream captured and stored in a suitable memory. This method is based on research of similar block only in the next frame or on the range of frames in the video sequence in question. The proposed process of tracking people in a single view from one static camera starts like any video surveillance system with background subtraction. Then, a Minimal Boundary Rectangle is used to surround the moving people in each frame. A selected block of 8×8 sizes is used to calculate the values of their entire joint blocks component in the following frames of the used video sequence. Finally, we search the similarity measure: The minimal of the Optimized Sum of Squared Difference and the Fast Fourier Transform Correlation. The same blocks are correlated in the posterior frames corresponding to the minimum error.