Wide baseline stereovision based obstacle detection for unmanned surface vehicles

Obstacle detection algorithm of a wide baseline stereovision system (WBSS) is proposed using deep learning for unmanned surface vehicle (USV). The WBSS is designed and constructed for a 7 m-class USV, which contains two fixed-focus cameras with a baseline length for 2 m. The obstacle detection algorithm contains three parts: firstly the sea-sky line is acquired by Hough transformation in an image, then the single shot multi-box detector (SSD) is adopted to detect the obstacles around the sea-sky line, and lastly F-KAZE algorithm is used for binocular image’s feature matching in a nonlinear scale space. The performance of the obstacle detection for the WBSS is tested at Jiaozhou bay in Qingdao city, and the sea experiment results prove availability and accuracy of the proposed technology.