基于自适应聚类算法的小群体检测与跟踪

We propose to detect and track small groups of individuals who are traveling together in surveiuance videos. Coherent groups are dynamically updated through merge and split events. To handle these challenges, we propose to discover groups by adaptive clustering (AGD). Experiments on challenging videos (FM dataset) which have complex motions and occlusions show that the proposed method based on trajectory-level similarity can correctly discover dynamic changes of groups. The effectiveness of the proposed approach is shown through comparison with classical methods.