Anomaly Detection in Crowded Scene

The use of cameras has increased, particularly in public areas like hospitals and city centers, mainly to detect and record anomaly events. Therefore, the automatic analysis of camera footage is important. This study proposes an algorithm that can detect anomalies in camera footage using optical flow map images and an autoencoder deep learning model, evaluating its performance with UCSD datasets. Separate training for each frame and determining a unique anomaly threshold for each test video enhanced performance. The performance of the proposed method was measured using the accuracy metric The proposed method achieved 81% accuracy on UCSD Ped-1 and 82% on UCSD Ped-2.