Dynamic anomaly detection and localization system for detecting abnormal events in crowded scene videos. This is an unsupervised model designed for the UCSD Pedetrian Dataset 2. The model is scene independent and can be easily extended to work on other video datasets. There is no need to explicitly define an anomaly. The training dataset consists of normal videos while the test set consists of normal and anomalous frames.
The work is presently under review at the Multimedia Tools and Applications Journal. The codebase will be updated to a working version once the review is complete.
For questions, please contact: medhini95@gmail.com