Anomaly Detection in Surveillance Videos

Automated anomaly detection is a useful task that can aid investigations and detect crimes. To this end, we present a model that can be used as a tool for anomaly detection in surveillance videos. Following an unsupervised approach, we use an autoencoder model trained to minimize the reconstruction error between the input and the generated output. We also augment the training of the auto-encoder with supervision in the form of user ratings per frame; higher user ratings reflect normal behaviour that the model is expected to faithfully reconstruct. On the other hand, lower rated frames are suspected to be anomalous. We analyze the output of the autoencoder on a standard dataset as well as two of our datasets that we have made public. We study the behavior of reconstruction error with and without supervision as well as the temporal coherence of the reconstruction error. Additionally, we use Grad-CAM to highlight potentially anomalous regions in the input. Finally, we discuss the problem of constructing summaries based on anomalous segments using heuristic approaches as well as a graph-theoretic formulation of determining a ranked list of maximum weighted cliques. We also make available in a single tool, our auto-encoder model as well as the anomaly summarizer.

[1]  Qingshan Liu,et al.  Abnormal detection using interaction energy potentials , 2011, CVPR 2011.

[2]  Abhishek Das,et al.  Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization , 2016, 2017 IEEE International Conference on Computer Vision (ICCV).

[3]  Jonghyun Choi,et al.  Learning Temporal Regularity in Video Sequences , 2016, 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[4]  Christophe Rosenberger,et al.  Abnormal events detection based on spatio-temporal co-occurences , 2009, 2009 IEEE Conference on Computer Vision and Pattern Recognition.

[5]  Yong Haur Tay,et al.  Abnormal Event Detection in Videos using Spatiotemporal Autoencoder , 2017, ISNN.

[6]  Nicu Sebe,et al.  Learning Deep Representations of Appearance and Motion for Anomalous Event Detection , 2015, BMVC.

[7]  Christophe Rosenberger,et al.  Abnormal events detection based on spatio-temporal co-occurences , 2009, CVPR.

[8]  Chen Shen,et al.  Spatio-Temporal AutoEncoder for Video Anomaly Detection , 2017, ACM Multimedia.

[9]  Shaogang Gong,et al.  A Markov Clustering Topic Model for mining behaviour in video , 2009, 2009 IEEE 12th International Conference on Computer Vision.

[10]  Randy C. Paffenroth,et al.  Anomaly Detection with Robust Deep Autoencoders , 2017, KDD.