マーカス ザーキイ,鈴木良太郎,朴錘一,井上正之
Video Rhythm and Motion Analysis
Abstract:Three different analysis techniques for video rhythm and motion analysis are presented. First, the spatial frequency analysis of movies is considered. Two sequences from different movies are compared subject to a variety of criteria with this method. A difference between the two sequences is characterized and various features are identified within each media. These features include scene changes, close-up movie shots, and long movie shots. Second, the principle component analysis of dance optical flow is considered. The optical flow of a segmented dance video is considered. A dance composition is segmented into parts that differ in terms of expression as displayed by the dancer. Each segment's respective motion data is reduced by principal component analysis (PCA). These principle components are compared to the principal components generated from a subjective experiment in which the subjects commented on parts of the dance composition. Lastly, a dance rhythm computation system is considered. A real-time system is developed in order to estimate a user's dance rhythm. The user views a skilled dancer and then attempts to duplicate this "training" dance rhythm. The system provides feedback to the user to indicate the closeness of rhythm attained. The rhythm calculation consists of a maximum entropy method (MEM) windowed frequency analysis of binary image differenced (BIN) data. Multi-user capability of the system is also explored.