Two and Three Dimensional Human Gait Modeling

Nowadays, human body modeling is becoming important in computer vision systems and gait has been found to be the most apparent feature than can be extracted from a distance. The ability for a video based surveillance system to automatically establish the model of the human model has various applications such as gait recognition and action recognition. However, the majority of existing analysis research is applicable to single view. Furthermore, some approaches are depending on a marker that is attached to a target which is not applicable for surveillance application. Therefore, the main objective of this research is to develop human gait model to extract features for surveillance application. The development of human gait model involves two major steps, a) 2D modelling from each individual view and b) estimation of 3D model from two combined views. Constructing a 2D model from the video data obtained from a single camera involves image pre-processing . Given the calibration information obtained using geometry based technique, the result of the 2D stick figure construction is used to estimate and reconstruct the 3D model in which the least number of two different views are needed.