자율 수중로봇을 위한 가중치 템플릿 정합 기반 물체 인식 및 추종

Abstract

In this paper, vision-based object detection and tracking techniques for autonomous underwater robots have been studied. Underwater vision has disadvantages of the limited detection range and the poor visibility; however underwater vision is still attractive for underwater robots in close range detections. In order to overcome the limitations and to make use of the advantages, in the previous works, we proposed the vision-based object detection and tracking methods using template matching and mean shift algorithms. But the performance was not satisfactory if the illumination condition changed. In this paper, the object detection method has been enhanced by employing the weighted sum of color-region-aided and feature-based approaches. We incorporated color information into the template matched area and the features of the template has been used to robustly calculate correlation coefficients. The performance of the proposed algorithm has been emonstrated through the water basin experiments using an underwater robot platform yShark made by KORDI.on is still attractive for underwater robots in close range detections. In order to overcome the limitations and to make use of the advantages, in the previous works, we proposed the vision-based object detection and tracking methods using template matching and mean shift algorithms. But the performance was not satisfactory if the illumination condition changed. In this paper, the object detection method has been enhanced by employing the weighted sum of color-region-aided and feature-based approaches. We incorporated color information into the template matched area and the features of the template has been used to robustly calculate correlation coefficients. The performance of the proposed algorithm has been emonstrated through the water basin experiments using an underwater robot platform yShark made by KORDI.