Abnormal Event Detection in Surveillance Videos Using Spatiotemporal Autoencoder
This research presents an efficient method for detecting anomalies in videos. Recent applications of convolutional neural networks have shown promises of convolutional layers for object detection and recognition, especially in images. However, convolutional neural networks are supervised and require...
| Main Author: | |
|---|---|
| Format: | Final Year Project / Dissertation / Thesis |
| Published: |
2019
|
| Subjects: | |
| Online Access: | http://eprints.utar.edu.my/3627/ http://eprints.utar.edu.my/3627/1/ESA%2D2019%2D1406689%2D1.pdf |
| _version_ | 1848885950092935168 |
|---|---|
| author | Chong, Yong Shean |
| author_facet | Chong, Yong Shean |
| author_sort | Chong, Yong Shean |
| building | UTAR Institutional Repository |
| collection | Online Access |
| description | This research presents an efficient method for detecting anomalies in videos. Recent applications of convolutional neural networks have shown promises of convolutional layers for object detection and recognition, especially in images. However, convolutional neural networks are supervised and require labels as learning signals. Hence, a spatiotemporal autoencoder architecture is proposed
for anomaly detection in videos including crowded scenes. The proposed architecture includes two main components, one spatial autoencoder for learning feature representation, and one temporal autoencoder for learning the temporal evolution of the spatial features. During training, the model is trained with only normal scenes, with the objective to minimise the reconstruction error
between the input video volume and the output video volume reconstructed by the learned model. After the model is trained, normal video volume is expected
to have low reconstruction error, whereas abnormal video volume is expected to have a high reconstruction error. By thresholding on the error produced by each testing input volumes, our system will be able to detect when an abnormal event occurs. The model is evaluated on four surveillance video datasets and compared using the area under ROC curve and abnormal event count.
Experimental results on UMN, Avenue, and UCSD benchmarks confirm that the proposed method can detect more abnormal events with lower false alarm rate than some state-of-the-art methods. The advantage of the proposed method is
that it is unsupervised — the only ingredient required is long video segments containing most normal events in a fixed view. Also, no feature engineering is required as the model automatically learns the most useful features from the training data. Further investigations will be carried out to improve the result of video anomaly detection by having human feedback to update the learned model for better detection and reduced false alarms. |
| first_indexed | 2025-11-15T19:30:44Z |
| format | Final Year Project / Dissertation / Thesis |
| id | utar-3627 |
| institution | Universiti Tunku Abdul Rahman |
| institution_category | Local University |
| last_indexed | 2025-11-15T19:30:44Z |
| publishDate | 2019 |
| recordtype | eprints |
| repository_type | Digital Repository |
| spelling | utar-36272019-12-17T04:36:06Z Abnormal Event Detection in Surveillance Videos Using Spatiotemporal Autoencoder Chong, Yong Shean QA75 Electronic computers. Computer science This research presents an efficient method for detecting anomalies in videos. Recent applications of convolutional neural networks have shown promises of convolutional layers for object detection and recognition, especially in images. However, convolutional neural networks are supervised and require labels as learning signals. Hence, a spatiotemporal autoencoder architecture is proposed for anomaly detection in videos including crowded scenes. The proposed architecture includes two main components, one spatial autoencoder for learning feature representation, and one temporal autoencoder for learning the temporal evolution of the spatial features. During training, the model is trained with only normal scenes, with the objective to minimise the reconstruction error between the input video volume and the output video volume reconstructed by the learned model. After the model is trained, normal video volume is expected to have low reconstruction error, whereas abnormal video volume is expected to have a high reconstruction error. By thresholding on the error produced by each testing input volumes, our system will be able to detect when an abnormal event occurs. The model is evaluated on four surveillance video datasets and compared using the area under ROC curve and abnormal event count. Experimental results on UMN, Avenue, and UCSD benchmarks confirm that the proposed method can detect more abnormal events with lower false alarm rate than some state-of-the-art methods. The advantage of the proposed method is that it is unsupervised — the only ingredient required is long video segments containing most normal events in a fixed view. Also, no feature engineering is required as the model automatically learns the most useful features from the training data. Further investigations will be carried out to improve the result of video anomaly detection by having human feedback to update the learned model for better detection and reduced false alarms. 2019 Final Year Project / Dissertation / Thesis NonPeerReviewed application/pdf http://eprints.utar.edu.my/3627/1/ESA%2D2019%2D1406689%2D1.pdf Chong, Yong Shean (2019) Abnormal Event Detection in Surveillance Videos Using Spatiotemporal Autoencoder. Master dissertation/thesis, UTAR. http://eprints.utar.edu.my/3627/ |
| spellingShingle | QA75 Electronic computers. Computer science Chong, Yong Shean Abnormal Event Detection in Surveillance Videos Using Spatiotemporal Autoencoder |
| title | Abnormal Event Detection in Surveillance Videos Using Spatiotemporal Autoencoder |
| title_full | Abnormal Event Detection in Surveillance Videos Using Spatiotemporal Autoencoder |
| title_fullStr | Abnormal Event Detection in Surveillance Videos Using Spatiotemporal Autoencoder |
| title_full_unstemmed | Abnormal Event Detection in Surveillance Videos Using Spatiotemporal Autoencoder |
| title_short | Abnormal Event Detection in Surveillance Videos Using Spatiotemporal Autoencoder |
| title_sort | abnormal event detection in surveillance videos using spatiotemporal autoencoder |
| topic | QA75 Electronic computers. Computer science |
| url | http://eprints.utar.edu.my/3627/ http://eprints.utar.edu.my/3627/1/ESA%2D2019%2D1406689%2D1.pdf |