In this paper, we discuss about using computer vision to identify human behaviour in security camera footage automatically. The majority of today's approaches calculate intricate characteristics by hand to use as the basis for classifiers. As a sort of deep model, convolutional neural networks (CNNs) are capable of operating on unprocessed data. However, at the moment, such models can only process 2D data. In this study, we create an innovative 3D convolutional neural network (CNN) model for recognising actions. By using 3D convolutions, this model is able to extract features from color, temporal, geometrical, and spatial dimensions effectively capturing the moving objects included in a sequence of consecutive frames. The input frame are used to produce various channels of data, and these channel are combined to construct the final visual features using the generated model. We propose regularizing the outputs using high-level characteristics and integrating the predictions of several algorithms to even further enhance the results. Models are then put to use in the real world, namely airport surveillance films, where they outperform state-of-the-art approaches in recognising human behaviours.