Multi-timescale Trajectory Prediction for Abnormal Human Activity Detection

12 Aug 2019  ·  Royston Rodrigues, Neha Bhargava, Rajbabu Velmurugan, Subhasis Chaudhuri ·

A classical approach to abnormal activity detection is to learn a representation for normal activities from the training data and then use this learned representation to detect abnormal activities while testing. Typically, the methods based on this approach operate at a fixed timescale - either a single time-instant (eg. frame-based) or a constant time duration (eg. video-clip based). But human abnormal activities can take place at different timescales. For example, jumping is a short term anomaly and loitering is a long term anomaly in a surveillance scenario. A single and pre-defined timescale is not enough to capture the wide range of anomalies occurring with different time duration. In this paper, we propose a multi-timescale model to capture the temporal dynamics at different timescales. In particular, the proposed model makes future and past predictions at different timescales for a given input pose trajectory. The model is multi-layered where intermediate layers are responsible to generate predictions corresponding to different timescales. These predictions are combined to detect abnormal activities. In addition, we also introduce an abnormal activity data-set for research use that contains 4,83,566 annotated frames. Data-set will be made available at Our experiments show that the proposed model can capture the anomalies of different time duration and outperforms existing methods.

PDF Abstract


Introduced in the Paper:

IITB Corridor

Used in the Paper:

ShanghaiTech CUHK Avenue

Results from the Paper

Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Anomaly Detection Corridor Ours AUC 67.12 # 2
Anomaly Detection CUHK Avenue Ours AUC 88.33% # 12
Anomaly Detection HR-ShanghaiTech Ours AUC 77.04 # 1
Anomaly Detection ShanghaiTech Ours AUC 76.03% # 14


No methods listed for this paper. Add relevant methods here