Research Behavior Analysis & Understanding
Action detection of TV viewer by object information and image feature
In this research, we propose a method to detect ‘ながら見行動’ which is thought to be important in estimating the degree of concentration on contents when watching TV. First, as the local feature quantity, motion information and angle information of joints based on joint position coordinates, object information around the hand, and image feature quantity independent of the joint position estimation result were calculated. Based on the method of Meshry et al., Each action is learned by Support Vector Machine using Bag of Feature.