A study on state recognition in wide area by aerial image analysis

Abstract

In recent years, the technologies of computer vision and robotics have been developed rapidly and used in many fields, especially on wide-area surveillance. Previously, several studies on visual surveillance have been executed. However, problems of narrow field of view and poor tracking capability still exist. To solve these problems, we propose a new visual control method to achieve automatic human tracking for flying robot. The method consists in integrating multiple modules to detect the moving direction of human automatically and tracking the subject in real time by performing autonomous control of the flying robot. To achieve automatic human tracking, it is essential to detect the moving direction of the subject accurately. Therefore, an algorithm to integrate human detection, human tracking, contour extraction, and moving direction detection and prediction is proposed in our study. In addition, flying robot is controlled by the correct moving direction of human. Evaluation experiments are carried out in simple environment and cluttered environment. Experimental results utilizing flying robot have shown the effectiveness of the proposed method.