Unusual Human Behavior Analysis Using the Deep Learning

The significance of abnormal event identification in video surveillance scenarios is receiving a lot of attention. There can be many purposes where these results can be utilized. It can be employed at railway stations, airports, malls, banks and hospitals for monitoring. Real-life events are frequently classified as normal or abnormal depending on their context. There are various approaches to detect the abnormities, from traditional method to deep learning. The difficulty of solving this problem is increased by the absence of samples with anomalous labels and it also increases model complexity. The present model tries to solve major issues like computational complexity and workload management. This study looks for real-world anomalies in CCTV footage, including abuse, assault, property damage, violence, blasting, and theft. These anomalies have a significant effect on human security and safety. Additionally, this work offers a cost effective deep learning method using CNN and LSTM for identifying abnormalities in acquired videos. This implementation is done with UCF crime dataset. The cascaded model gives 98.5% accuracy for detection six unusual activities.