| Popis: |
Children's care and education is an eternally important element of society. In nursery schools, because children are active and sensitive, it is always difficult to effectively analyze the natural behavior of children if an observer (a person holding a camera) exists. In this paper, we collect a large amount of video data from different nursery schools by using stationary cameras without attaching motion sensors on children. We construct a system for children's behavior analysis based on human tracking from the videos in nursery schools. To achieve effective human tracking in videos of nursery schools, we combine multiple computer version technologies such as person detection, object tracking, and person re-identification. For each technique, we comprehensively compare various methods and identified the best one. With the system, we analyze some behavioral patterns of children in nursery schools. Specifically, we identify the popular areas in the room, the amount of exercise, and the gregariousness of each child. |
