Automatic multiple human tracking using an adaptive hybrid GMM based detection in a crowd

Abstract

For a visual surveillance in a crowd, multiple human tracking is essential and of course a challenging task. Real-world applications require multiple cameras to capture crowd scenes so that a keen tracking is observed. Automatic tracking in a crowded environment is very important criteria for the surveillance. Accurate and real-time tracking in a crowd, the number of people present in the public places and shopping mall are some of the vital information for monitoring traffic violations. To provide human safety and security, surveillance like theft prevention and automated checkout provides the necessary consumer information to the managers. The conventional tracking algorithm does not handle the complex background, multi-view points, various illumination changes and severe occlusion occurring in a crowd. The above problem can be effectively handled by using the proposed Adaptive Hybrid Multiple Human Tracking (AHMHT) method. The proposed work utilizes the Adaptive Hybrid Gaussian Mixture Model (AHGMM) (Karpagavalli and Ramprasad, International Journal of Multimedia Tools and Application 76(12):14129–14149, 12) detected output, so that, the proposed algorithm tracks all the blobs in each frame on the basis of motion information along with the width and height information of exact blob. The experimental results demonstrate that the proposed method performs well compared to other methods. The multiple human tracking rates are improved with maximum of 91% using the proposed frame work compared with other methods. The proposed method is efficient in terms of computational time (CT) using an adaptive hybrid tracking.