Аннотация:In the article we describe a problem of a fully automatic people tracking in surveillance video recorded by a stable camera. We propose a development of Benfold et al tracking-by-detection algorithm [1]. We introduce a filtering of person detector results and a modified method of tracklet’s position estimation. Also we use the Flock of features and the normalized cross-correlation, as visual tracking algorithms, to construct tracklets. Our experiments reveal that all proposed modifications allow producing more robust trajectories than trajectories produced by basic algorithm.