Research Behavior Analysis & Understanding
Behavior Recognition Method of Passenger in Vehicle Based on Global Structure of Human Body Part Location
In the proposed method, we first estimate the location of the human body part of the driver using depth learning using two modals, an IR image and a depth image as input. Also, based on the global structure of the position of the obtained part, it is classified into a characteristic posture expressed in a certain behavior and a common posture expressed in a plurality of behaviors. Further, the probability density for each recognition target category of the obtained key-pose is integrated in the time direction, and predetermined action is recognized by increasing the accuracy. Here, a random forest was used for key-pose extraction.