Detection of Abnormal behavior in Dynamic Crowded Gatherings

Abstract

People gather for parades, sports, musical events, and mass gatherings for pilgrimage at religious places like Mecca, Jerusalem, Vatican, etc. Most often, these mass gatherings lead to crowd disasters. In this research, a new automated algorithm for the Detection of Abnormal behavior in Dynamic Crowded Gatherings (DADCG) is proposed that has reduced processing speed, sensitivity to noise, and improved accuracy. Initially, the temporal features of the scenes are extracted using Motion History Image (MHI) technique. Then the Optical Flow (OF) vectors are calculated for each MHI image using Lucas-Kanade method to obtain the spatial features. This Optical flow image is segmented into four equal-sized blocks. Finally, a two dimensional histogram is generated with motion direction and motion magnitude for each block. Stampede and congestion areas can be detected by comparing the mean value of the histogram of each segmented optical flow image. Based on this result, an alarm may be generated for the security personnel to take appropriate actions.