Hybrid Approach For Real Time Crowd Activity Identification Using Segmentation

Crowd investigation turns into the most dynamic research in PC vision, in the shrewd video reconnaissance zone. In this research, an online solution is implemented to detect crowd abnormalities, which can be caused by panic. After calculating optical flow and calculating magnitudes using the Farneback method, an activity map is created using multiple frames to see the flow persistency overtime. The activity map is used to create two metrics, TOV (Temporal Occupancy Variation) and entropy, respectively. In this research, the measurements alongside an edge to the group are utilized. Subtleties will be examined in the accompanying segments. The movement directions are centered to watch the group and optical stream strategies are utilized to gain the streak lines and faculty directions. Also, the exhibition measurements are computed like accuracy FTR, TPR, and ROC.