Enhanced Human Interaction Recognition Framework using Pyramid Matching and Deep Neural Network

The interaction of humans on images has been of interest in computer vision and pattern recognition. In fact, interaction recognition has become fertile ground for research in virtually all areas of research. Thus, proposing to use a Deep Neural Network to design an activity recognition system in that case. However, at the initial stages of the system, the integration of hue, saturation, intensity (HSI) color transformation brings clarity of the video frames. Additionally, filtering for noise interferences are used to minimize noise. Multiple Object Tracking (MOT) and statistical methods are used to extract silhouette. In the extraction procedure, we use the Texton maps, ORB (Oriented FAST and Rotated BRIEF), and Spatial pyramid matching (SPM) Method. Using Gray Wolf algorithm, we discriminate features and then identify the most meaningful independent content that describes the data structure. Finally, the DNN takes the last feature and dumps it into the human interaction zone. This approach is tested using the SBU Interaction dataset and achieves a recognition rate of 87% using the experimental method.