Crowd count in low resolution surveillance video using head detector and color based segementation for disaster management

In any public gathering, to ensure the safety of the people during or after any kind of natural or man-made calamities, counting the number of people in a crowd is of paramount importance. In cases of unpredictable environments where occlusion, shadows, varied illumination and multiple objects in the background persists, many algorithms fail to give out accurate results. In this paper, to overcome this predicament, a color based segmentation and a generic head detector has been used. First, the color based segmentation is implemented to detect the skin tone for person identification. After which, the entire scene will be divided into four quadrants and a generic head detector is used to count the number of people in each quadrant. The advantages of two methods are to increase the accuracy and make the counting system fail proof. The need for dividing the scene into four quadrants is to provide location based density estimation so that immediate help can be sent in case of any calamity. Based on the developed approach, a robust counting system for fast real time scenes is developed. Extended experimental results illustrate the effectiveness of the proposed algorithm.