Anomaly detection using improved background subtraction
Yunus Can Bilge, Fikret Kaya, Nazlı İkizler-Cinbiş, Ufuk Çelikcan, Hayri Sever · 2017
Detection of abnormal state (anomaly) is one of the topics that is frequently studied in the field of computer vision. In this study, we aim to investigate the combination of an unsupervised and discriminative anomaly detection method with a background subtraction technique. In this context, when the discriminative anomaly detection technique is used in conjunction with the recommended background subtraction method, it is seen that more efficient and meaningful results can be obtained. With the proposed method, the noisy features that extracted from the background are avoided, and as a result the detection performance is increased. Experimental tests on a dataset demonstrate that, the proposed method effectively increases the anomaly detection performance.