Human Action Recognition System using Good Features and Multilayer Perceptron Network

摘要

Human action recognition involves the characterization of human actionsthrough the automated analysis of video data and is integral in the developmentof smart computer vision systems. However, several challenges like dynamicbackgrounds, camera stabilization, complex actions, occlusions etc. make actionrecognition in a real time and robust fashion difficult. Several complexapproaches exist but are computationally intensive. This paper presents a novelapproach of using a combination of good features along with iterative opticalflow algorithm to compute feature vectors which are classified using amultilayer perceptron (MLP) network. The use of multiple features for motiondescriptors enhances the quality of tracking. Resilient backpropagationalgorithm is used for training the feedforward neural network reducing thelearning time. The overall system accuracy is improved by optimizing thevarious parameters of the multilayer perceptron network.