Appearance-based person identification algorithm to search for persons using a flying robot

This paper proposes a person identification algorithm that can be used when searching for persons from multiple images captured by flying robots. The images taken by such robots have low resolution and contain various viewing angles of people because they fly at a safe distance from and often move around people. The proposed person identification algorithm is based on appearance and can identify a person from low-resolution images of the upper body. In appearance-based person identification, the accuracy is low when a single image is used. Therefore, the accuracy is improved by accumulating the score. Experimental results obtained using 40 subjects show that the identification accuracy based on images having different view angles is better than that based on images having similar view angles. Further, the identification accuracy increase even when rear images of the person that are barely distinguishable from those of another person are added to the input images. In addition, it can identify persons even when the size of the upper body image is as small as 15 pixels. We performed the identification experiment using images taken by an aerial-robot while it was actually flying and proved the effectiveness of the algorithm.