A robust and enhanced approach for human detection in crowd

Abstract

In this paper we have presented and enhanced methodology for robust detection of human in videos. Our research covers most of the limitations for detection of human in crowded places like detection of non-human objects and large human shadow as humans. The proposed technique is based on hierarchal structure consisting of three phases. Firstly, segmentation of moving objects is done using Gaussian Mixture model. Secondly, shadow removal technique is applied to avoid detection of large human shadows as human. Finally, human detection is achieved by applying human detection algorithm [3] on shadowless segmented images. Experiments are performed on different videos having single and multiple humans in indoor and outdoor scenes and videos under different illumination producing large and small shadows. This paper also presents comparative results of our methodology with existing techniques and the results clearly proved that the proposed technique outperforms the existing techniques and this is proved by producing comparative results.