Complex human activity searching in a video employing negative space analysis

In this paper a region based technique is proposed to search complex human activity in a video. Most of the current activity recognition systems deals with single action type but in this system, by taking meaningful composition of different types of human actions, complex human activity is constructed which is searched in a video sequence. To extract features, this system employed the surrounding regions of the human silhouette which is termed as negative space according to art theory whereas other region based methods work on the silhouette of the person. Negative space based features have the ability to describe human poses with simple shapes implying simple description of poses. Moreover, negative space features are less sensitive to noise, overcome some limitation of silhouette based methods such as leaks or holes in the silhouette, robust to partial occlusion, shadows and different clothing. The system consists of hierarchical processing of background segmentation, region partitioning, extraction of shape based features, finding matching score by Dynamic Time Warping and speed calculation. The system is compared with state of the art method and offered significantly improved performance over different complex activities.