Panic detection by regional velocity changes in crowded areas from surveillance video

Common use of surveillance cameras and processing data gathered from them with only human power makes occur some difficulties about time and workload. In this work, a convolutional neural network based system is designed to detect panic situation with evaluating velocity changes of people by regions. Convolutional neural network for human detection; Kalman filter, Hungarian algorithm and convolutional neural network for extracting trajectory is used. The determination of the panic condition is made by evaluating the difference between the short and long time average of the speed of the people in each region. Unlike existing systems, by evaluating velocity changes on display locally, false positive results are reduced so that panic situation is detected more accurately.