Waterline and obstacle detection in images from low-cost autonomous boats for environmental monitoring

Waterline detection from images taken by cameras mounted on low-cost autonomous surface vehicles (ASVs) is a key process for obtaining a fast obstacle detection. Achieving an accurate waterline prediction is difficult due to the instability of the ASV on which the camera is mounted and the presence of reflections, illumination changes, and waves. In this work, we present a method for waterline and obstacle detection designed for low-cost ASVs employed in environmental monitoring. The proposed approach is made of two steps: (1) a pixel-wise segmentation of the current image is used to generate a binary mask separating water and non-water regions, (2) the mask is analyzed to infer the position of the waterline, which in turn is used for detecting obstacles. Experiments were carried out on two publicly available datasets containing floating obstacles such as buoys, sailing and motor boats, and swans moving near the ASV. Quantitative results show the effectiveness of the proposed approach with 98.8% pixel-wise segmentation accuracy running at 10 frames per second on an embedded GPU board. •A pixel-wise segmentation is used to separate water and non-water regions.•The water/non-water mask is analyzed to infer the position of the waterline.•Waterline information is used to reduce the search space for detecting obstacles.