Unsupervised Identification of Coherent Motion in Video
Luciano Silva da Silva, Jacob Scharcanski · 2007
The identification and classification of motion patterns in point trajectories has been an important issue in understanding and representing dynamic scenes. This paper proposes an unsupervised approach to identify coherent motion in video. Instead of producing a spatio-temporal segmentation of the raw data, the proposed method analyzes point trajectories along the video sequence to identify sets of points that move coherently. This new way of extracting motion information from videos potentially can be used in different areas of image processing and computer vision.