The aim of this work is to track humans in a known environment. The input data are pre-segmented video streams captured with a fixed camera.  In this paper we propose a new approach to the problem of simple hypothesis non-delayed human tracking.  Furthermore the approach handles simultaneously and in a unified manner the actual similarity matching and the associated real-world problems of entries, exits, occlusions and false detections (noise). Results on visual surveillance applications for security in metro stations , security in banks and monitoring in offices are shown. Quantitative results of this method in terms of time performance are presented.