Detecting Dominant Motions in Dense Crowds

Crowded Scene Analysis: A Survey

“…Besides, to evaluate the generated paths, we first cluster each detected path shown in Fig. 7 into a smooth dominant trajectory using the method proposed in [69], then we compute the similarity between the dominant trajectory and the labeled path. A distance measure of longest common subsequences (LCSS) [76] is used.…”

“…Cheriyadat et al [69] used a distance measure for feature trajectories based on longest common subsequence (LCSS) [73]. The method begins with independently tracking low-level features using optical flow, and then clusters these tracks into smooth dominant motions.…”

Crowd Analysis Using Computer Vision Techniques

“…In general, the temporal information is used to estimate the behavior of a crowd in a given environment, such as main directions [40], velocities [5], and unusual motions [25], [35], [8]. A great variety of approaches were proposed in past years to deal with crowd analysis and understanding that could involve researchers from several areas.…”

“…Cheriyadat and Radke [40] proposed an approach for clustering a set of low-level motion features into trajectories, similarly to [34] and [35], but using additional rules in the clustering process, such as the dominant movements that are computed based on the longest common subsequences. However, while the goal in [34] and [35] was to identify each member in the scene based on motion cues, the main goal in [40] was to extract dominant motion patterns in a crowded scene.…”

“…However, while the goal in [34] and [35] was to identify each member in the scene based on motion cues, the main goal in [40] was to extract dominant motion patterns in a crowded scene. Then, individual motions not coherent with dominant flows could be highlighted and marked as potential unusual behavior.…”