Extreme Contextual Multi-Anomaly Identification And Tracking From Aerial Down Shot View
Indrajit Kar, Anindya Chatterjee, Jacob Vishal, Sakshi Tyagi · 2023
The total performance of tracking has improved because of the presence of several MOT (multi-object detection) systems. However, it has problems with motion fluctuations, puzzling visual changes, object detection, and tracking in crowded areas. Due to the imbalance between normal and abnormal data points, the video anomaly detection problem has long been regarded as the most difficult in visual anomaly tracking and identification. The objects of interest in the experimental study briefly appear throughout the entire tape, making it an uncommon occurrence. In addition to spatial and temporal data, video anomaly detection also considers contextual anomalies, which are anomalies resulting from the local context. The method of the Authors for detecting anomalies entails tracking many anomalies until they disappear. The down shot angle and abrupt change in video color in the Experiments set them apart from all previous video anomaly detection problems. The authors create several models that are listed below making this a novel Video anomaly detection and tracking paper.