An innovative volume based video feature extraction technique

Based on the video frames, a spatial-temporal volume data structure represents more flexible processing methods than traditional 2D sequential images approach in computer vision. This paper describes the data structure of the spatial-temporal volume and the feature volume coming from the original video data. A compressed volume structure called a feature video is presented showing how a feature volume can be build based on the sequential frames. The existent 3D volume processing methods such as slice processing, 3D filters and CFD methods are also introduced in this paper. As a practical application, human gait analysis based on volume slice processing is described. This includes the video capture, volume organized and the slice feature extraction. The result of experimental data shows the different features between different behaviors of gait.