Identifying People Using Body Sway in Case of Self-Occlusion

We segment the head region in overhead images and measure body sway in local blocks. Even when parts of the body are hidden by the body itself, small head movements can be used to identify people.

People sway slightly even while standing still, and these movements can differ between individuals. In an overhead view, however, one part of the body may hide another. This is called self-occlusion. It can reduce the accuracy of a method that uses the whole-body image to identify a person.

We focus on movements of the head instead of using the entire body. First, we separate the head region from the rest of the image and create a silhouette showing its shape. We then divide that region into small blocks and measure how each block moves over time. The resulting description captures where and how the head sways, providing information for distinguishing individuals.

In experiments, the proposed method improved identification performance over the existing method used for comparison. The results show that small head movements can be useful for identifying people when parts of the body are hidden in an overhead view.

• Takuya Kamitani, Yuta Yamaguchi, Masashi Nishiyama, Yoshio Iwai,
  Identifying People Using Body Sway in Case of Self-Occlusion,
  Proceedings of International Workshop on Frontiers of Computer Vision (FCV), OS1-4, pp. 1 - 13, February 2020.
• 神谷 卓也, 山口 優太, 西山 正志, 岩井 儀雄,
  天井カメラを用いて観測された身体動揺における自己遮蔽の影響を考慮した人物対応付け,
  電気学会論文誌C, vol. 140, no. 6, pp. 629 - 637, June 2020.

• Body Sway in Case of Self-Occlusion