A method was proposed to detect so called "texting while walking" from positions of human body 25 keypoints estimated by the OpenPose for a person in a video scene. The method was based on the fact that movements of human joints had some spatial correlation structures. Walking behaviors with or without using a smartphone were separated by the difference between the correlation structures. We evaluated the difference using a canonical correlation analysis applied to two covariance matrices

more »

... ated from pose vectors for both walking behaviors, respectively. The canonical score well separated the texting while walking from the normal one. The proposed method was successfully applied to actual video scenes for three persons' three types of walking behaviors.