Anomaly Detection
Anomaly detection in videos refers to the identification of events that do not align with the expected behavior. This paper is a pioneering work that leverages the difference between a predicted future frame and its ground truth to detect an abnormal event.
There are mainly 3 elements.
- Image Generation
- Image Discrimination
- Optical Flow Estimation
Image Generation
UNet to generate the image of the next time frame using all of the previously observed images.
Image Discrimination
GAN network to distinguish the label image from the generated image from UNet.
Optical Estimation
Use the image of the current time frame and the image of the next time frame as a pair input for FlowNet to estimate optical flow.
Reference
Future Frame Prediction for Anomaly Detection – A New Baseline
