COMPARATIVE ANALYSIS OF CONVOLUTIONAL NEURAL NETWORK MODELS APPLIED IN HUMAN BEHAVIOUR RECOGNITION

Abstract

Human behaviour recognition is a crucial area of scientific research in the science of computer vision that has significant applications in a variety of industries, including intelligent surveillance, smart homes, and virtual reality. Traditional manual approaches have a hard time meeting the demands of high recognition accuracy and applicability in the contemporary complicated environment. Deep learning's arrival has opened up new avenues for behaviour recognition research. The major focus of this paper is behaviour recognition using convolutional neural networks (CNN). Before discussing and analysing the classical learning methods and deep learning methods of behaviour recognition, the research context and importance of behaviour recognition are first introduced. Based on the convolution neural network designed for the specific human behaviour in public areas, we develop a series of human behaviour recognition systems. In order to extract behaviour or characters of the body, the video and images of human behaviour data set will be processed. Then the planned convolution neural network is trained with the training data sets, and the deep learning network is built. Finally, using the developed network model, the numerous sample behaviours are categorized and recognized, and the recognition outcomes are then evaluated. The findings demonstrate that CNN is capable of studying human behaviour models automatically and recognizing human behaviours without the need for manually annotated training.