Human tracking in multiple non-synchronous cameras

Abstract

In this project, we will study the human tracking from multiple non-synchronous view video sequences. Some techniques are used to retrieve or track the human motion from the video sequence to analyze the human motion or do human tracking. The track of the human motion is located on the x-y plane. In the past years, there are many researches on motion tracking in single view video sequence. However, there are some problems, such as occlusion by another people or self-occlusion. Recently, some researches have adopted the multiple-viewpoint approach aiming to avoid those problems mentioned before. In order to separate the human from the surrounding, a simple background subtraction is applied to obtain the human. After locating the position of the human from the image extracted from the video sequence, the track of the human is predicted by the Kalman filter. Also, a co-planar calibration is applied to project the 2D coordinate from an image to a 3D world coordinate. Video sequences are captured to test the algorithm.