Tracking Humans Based on Interest Point over Span-Space in Multifarious Situations

Abstract

Tracking humans in complex surroundings is a very vital, exciting and red-hot topic since last few decades. There are lots of opinions and approaches proposed and developed by different researchers to prove that their methodology is comparatively superior to others. Human activity can be tracked based on the appropriate motion of human being. This motion is segregated into two categories as global motion and limb motion. We assumed that human may carry something with him or her. However, to apply our method everything is considered as object in our proposed model. The objects are being tracked, segmented and subsequently are being classified. The global-parts of this human motion approach (HMA) are being tracked by using ellipsoid shape of human models. However, based on the tracked object we tried to establish the performance of matching and recognition of human object methods (HOM). Here we have considered interest-points as a tracking parameter of the associated image descriptor. The interest-points measured in scale based on the shape of span-space of tracked-image-object (TIO) and they should fall into one of the category (global motion or limb motion). In this paper we propose the way to detect and identify interest-points by using generalized detectors. Subsequently we shall build sparse-matrix set of interest-points to compute image descriptors for human-image-matching (HIM). To detect the scale of interest-points based on span-space; however, we have used sophisticated statistical and mathematical approaches so that we can get more accurate experimental results. Our experimental results show that it is successfully applied in case of small number of people moving together accompanied by occlusion and shadow or reflection effect. However we have also tried to implement the different modes of human activity (e.g., standing, playing, talking, walking, running, etc.). There are some constraints for model of camera and the assumptions required for geometric shape of ground plane. Robust results show the superiority of our applied methodology.