Alleviating Undesired Distance Effect in Spatio-Temporal Based Video Anomaly Detection

Anomaly detection in videos plays a vital role in a surveillance system by identifying events that deviate from normal behavior. A key challenge in Video Anomaly Detection (VAD) is detecting unusual activities that occur at far distances from the camera. To address this issue, we have proposed a novel Weight Map to enhance the detection of distant anomalies. We have evaluated our proposed method on two commonly used datasets UCSD Ped1 and Ped2. Experimental results demonstrate considerable performance improvements, achieving an AUC of 96.04% on Peds2 and 81.38% on Peds1, proving the effectiveness of the proposed weight map in identifying anomalies, particularly those that occur far from the camera.