Unusual Event Detection in Surveillance Video Using Transfer Learning

Unusual event detection in surveillance video is one of the active research area in computer vision. In this paper, we study the performance of different classifiers using a two-stream CNN architecture. The CNN along with different classifiers have been used to detect unusual event in a surveillance video. The two-stream two-dimensional convolutional network pretrained on the ImageNet database is employed to extract features from video frames. The extracted features are used to classify the video frame into one of the two classes. Different classifiers including SVM, K-NN, RBFN, Naive Bayes, Logistic Regression, K-means clustering have been utilized for this purpose. It has been observed that SVM gives better performance as compared with other classifiers.