Anomaly Detection Using Transfer Learning in CNNs


The demand for greater security measures in crowded environments for monitoring and protecting operations has made video anomaly detection significant study area in computer vision. This is due to the fact that the detection of anomalies in video surveillance systems has increased, making it one of the leading focus areas in the current field of research. In this study, we offer a method for identifying out-of-the-ordinary anomaly behaviour in video footage of crowded settings. The proposed approach uses the three CNN architectures of AlexNet, ResNet and VGGNet. Fine tunes the architectures by using conjugate gradient optimization. The transfer learning approach is followed for the classification. The proposed method saves time by not performing the training part again and combines the results from the three architectures. Random forest and Softmax classifier perform the classification. The proposed model faresbetter than some o

Full Text Attachment