Vision-based Detection of Humans on the Ground from Actual Aerial Images by Informed Filters using Only Color Features

In this paper, we construct a novel sensor network stem that can measure real-time vital signs of the human body during exercise even at high speeds and in crowded regions. The sensor network estimates locations of sensor nodes using image processing to extract locations of humans wearing sensors. This paper evaluates the accuracy of human detection by informed-filters using only color features for actual aerial images. To carry out the evaluation, a novel dataset composed of actual images captured using a camera mounted on a drone was created. Experimental results show that significant accuracy can be achieved by the detector. In addition, the number of weak classifiers in a strong classifier can be reduced to 125 without significant degradation of the detection accuracy.