Human Behaviour Classification for Video Surveillance Using CNN

The surveillance systems were introduced to assist the security officers to prevent crimes. The addition of deep learning technologies to such systems has led to a revolution, converting them into smart video surveillance, which could do automatic human behaviour classification. This paper presents a transfer learning approach by using VGG16 as a feature extractor and convolutional neural network as a classifier. The report mainly focuses on strategies for reducing the input features and training the model to classify three human behaviours, namely, run, walk and wave and comparing the results based on those approaches. The experimental results show that the proposed algorithm of removing the irrelevant features, allows the model to focus only on the essential details and gives better results as compared to earlier methods. An increase in accuracy can be seen with the removal of each irrelevant input feature.