Context Aware Crowd Tracking and Anomaly Detection via Deep Learning and Social Force Model

The world’s expanding populace, the variety of human social factors, and densely populated environment make humans feel uncertain. Individuals need a safety officer who generally deals with security viewpoints for this frailty. Currently, monitoring techniques are time-consuming, work concentrated, incapable. Therefore, autonomous surveillance frameworks necessary modern day since they able to address these problems. Nevertheless, hardships persist. central concerns incorporate detachment foreground from scene understanding contextual structure efficiently identifying unusual objects. In our work, we introduced novel framework tackle difficulties by presenting semantic segmentation technique separating object. Super-pixels generated using an improved watershed transform then conditional random field is implemented obtain multi-object segmented frames performing pixel-level labeling. Next, Social Force model extract via fusion chosen particular histogram optical stream inner force model. After computed force, multi-people tracking performed three-dimensional template association percentile rank non-maximal suppression. categorization deep learning Feature Pyramid Network. Finally, considering environment, Jaccard similarity utilized decision abnormality detection identify objects scene. invented verified through rigorous investigations, it obtained efficiency 92.2% 89.1% over UCSD CUHK Avenue datasets. However, 95.2% 93.7% accomplished datasets, respectively.