Instant coherent group motion filtering by group motion representations

Na Li, Yun Zhang, Wenhan Luo, Ning Guo

Abstract: Highlights • Propose two types of instant motion representations of group motion consistencies. • An instant group refining framework and instant group motion consistency filtering strategy are proposed based on the instant motion representations. • A spatial distance robust group detection method is proposed. Abstract Group motion direction, which is composed of instant consistent motions, is an interesting property in group dynamics. Understanding the instant group motion direction is a key component of crowd analysis and has a wide range of applications, especially for improving group detection in crowd. In this study, independent from group detection methods, we propose an instant group motion refining framework based on group motion direction. We show that instant group motion direction can be systematically quantified by intra-group motion consistency. This is achieved by proposing two instant group motion representations in terms of velocity consistency and angle consistency. These group motion representations provide a new way to derive the instant group motion direction, which plays an important role in improving the performance of state-of-the-art group detection methods. Besides, to improve traditional group detection methods, a novel group detection method is proposed as well. Extensive experiments on a variety of public scene video clips demonstrate that both group motion consistency representations are not only useful but also necessary for instant group coherent motion filtering.