Anomaly detection in Video Feeds
The official repository of our ML Project Anomaly detection in Video Feeds.
Team Name
Team Gabru
Team Members
Aishik Chakraborty 13CS30041
Ashish Sharma 13CS30043
Chinmaya Pancholi 13CS30010
Jatin Arora 13CS10057
Jeenu Grover 13CS30042
Prabhat Agarwal 13CS10060
Presentation
The presentation can be found here (Slides, pdf)
Approaches
1. Using Optical Flow: Extracting spatial-temporal features using optical flow and training classifiers for detecting anomaly
2. Using AlexNet: Alexnet contains eight learned layers, five convolutional and three fully-connected layers. Pretrained Alexnet is applied to each frame of the video. Then we take the output of the fc7 layer which gives us a 4096 dimensional vector.
3. Time Series Analysis: Standard statistical techniques for anomaly detection in the feature space we obtain from AlexNet.
4. Topic Modelling: A generative model of typical behavior is learned using good discriminative features, and then abnormal behaviors (outliers) are detected and classified as those that are badly explained by the learned model.
References
- Rousseeuw, P. and Leroy, A.: 1996, Robust Regression and Outlier Detection. John Wiley & Sons., 3 edition
- Torr, P. H. S. and Murray, D. W.: 1993, ‘Outlier Detection and Motion Segmentation’. In: Proceedings of SPIE
- Fawcett, T. and Provost, F. J.: 1999, ‘Activity Monitoring: Noticing Interesting Changes in Behavior’. In: Proceedings of the 5th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. pp. 53–62