Identification of photo-taking behaviors using optical flow vector

The rate of smartphone ownership has significantly increased all over the world year by year. According to the statistical data by Japanese government, more than 90% of people aged between 20 and 30 own smartphones as of 2017. Smartphones are very useful and it is easy for people to communicate with each other, taking pictures and sharing the pictures on SNS (Social Networking Sites). However, there exists an important social problem related to taking pictures, namely, unintended appearance in photos. When someone is taking a photo in a public place, other people may appear in the photo unintendedly due to the lack of photographer’s moral, resulting in a privacy risk of the photographed persons. To avoid such a situation, most of existing studies perform image processing to the photo image, e.g. superimposing pixelated or blurred images around the faces. This is a passive approach for the photographed persons conducted at the photographer side, and thus, there still exists a privacy risk. In this research, an active approach conducted at the photographed person side is proposed, aiming at detecting photo-taking behaviors by smartphone. In the proposed approach, the photographer’s behaviors, which show someone is about to take a photo, are focused on. It is assumed that a photographed person (user) wears a small camera like a “life log camera” and monitors his/her surroundings. The final goal of this research is to detect whether the person is about to take a photo or not, based on the video data analysis. In this paper, we analyze the characteristics of photo-taking behaviors using Optical Flow technique, referring to the movement of arms and/or hands of human. The result of evaluation experiments reveals an interesting feature distribution and shows that the detection accuracy of photo-taking behaviors is 67%.