2A1-Q04 Unsupervised Generation of Binary-Tree and Supervised Learning of Linear Classifier for Realtime Statistical Recognition of Human Behaviors(Informative Motion & Motion Media)

Abstract

This paper proposes motion recognition method by constructing binary tree database and searching it with Support Vector Machine(SVM). We defines the kernel function for the motion feature space. Kernel method, which defines the inner product in the feature space, makes it possible to analyze Hidden Markov Model(HMM) of human motion data and construct SVM in the motion feature space. Motion binary tree has been used for hi-speed motion recognition but it is only based on Unsupervised learning , or non-human sense. So, by integrating SVM into the binary tree, it enables to search the binary tree with Supervised learning. We validate proposed recognition method with 1500 captured motion data.