Coherent Motion Detection with Collective Density Clustering

Novel Group Detection and Analysis Method Based on Automatic and Fast Density Clustering

“…State-of-the-art methods [3,6,7] have achieved group detection through tracklet analysis. Shao et al [3] proposed the novel Collective Transition (CT) prior to capture the underlying dynamics of a group and devised a set of visual descriptors to quantify the universal properties of groups in a crowd.…”

“…Shao et al [3] proposed the novel Collective Transition (CT) prior to capture the underlying dynamics of a group and devised a set of visual descriptors to quantify the universal properties of groups in a crowd. Wu et al [7] developed the Collective Density Clustering (CDC) approach, and Zhou et al [6] presented Coherent Filtering (CF) to recognize coherent motion. However, the above methods are unfeasible for real-time application or require manual intervention.…”

“…Because the accuracy of heavily depends on the threshold , we propose an automatic method to handle this inevitable problem by using entropy without depending on subjective experience [7,17]. We were inspired by the Shannon entropy; in information theory, entropy is often used to quantify the uncertainty of a probability distribution when a specific event is observed.…”

“…Understanding group dynamics is scientifically important and practically useful for crowd behavior analysis [3,4]. However, a specific definition for a group in a crowd has rarely been given, even though groups have been extensively studied in the socio-psychological and biological fields [5][6][7].…”

“…Other methods [18,19] that use a motion flow field are limited to specifying the clustering number manually. The most recent methods [17,7] require a factor given manually, which needs field experience. Shuai et al [2,11] tried to analyze group attributes but could only handle the problem when the groups are stationary.…”