Abnormal Activities Recognition

This week, I'll review a paper about abnormal activities, which currently become my research topic for my disertation. I would to review a famous paper entitled with "Real-world Anomaly Detection in Surveillance Videos", which published in CVPR 2018. Aside with proposing a new method for crime detection, this paper also propose a new benchmark dataset for this problem.

This paper considers normal and anomalous videos as bags and video segments as instances in multiple instance learning (MIL), and automatically learn a deep anomaly ranking model that predicts high anomaly scores for anomalous video segments. Furthermore, we introduce sparsity and temporal smoothness constraints in the ranking loss function to better localize anomaly during training.

The flow diagram of the proposed anomaly detection approach. Given the positive (containing anomaly somewhere) and negative (containing no anomaly) videos, it divides each of them into multiple temporal video segments. Then, each video is represented as a bag and each temporal segment represents an instance in the bag. After extracting C3D features for video segments, it trains a fully connected neural network by utilizing a novel ranking loss function which computes the ranking loss between the highest scored instances (shown in red) in the positive bag and the negative bag.