HUMAN ACTION RECOGNITION NETWORK CLASSIFIER LEVERAGING SVM-DIRECTED MACHINE LEARNING

Abstract

To ensure public safety, successful surveillance video analysis requires both an understanding of human behavior and the ability to recognize human actions. However, because they require a large amount of parameters, existing approaches like neural networks and three-dimensional convolutional neural networks (CNN) have computing challenges. To address these challenges with regard to efficient human action detection, a specific residual CNN based on directed acyclic graphs was created in this work.  Effective implicit representational acquisition of human movements is rendered simpler by applying the suggested method's novel pipeline for generating spatial motion data from raw video inputs. The implicit representations, are used as deep learning protocols and are maintained in a fully linked layer. Action recognition is then accomplished by passing these suggested approaches into the Support Vector Machine (SVM) classifier.  Simulations were conducted using MATLAB software on widely used behavior recognition statistics to assess the created SVM algorithm.