Deep learning waterline detection for low-cost autonomous boats

Abstract

Waterline detection in images captured from a moving camera mounted on an autonomous boat is a complex task, due the presence of reflections, illumination changes, camera jitter, and waves. The pose of the boat and the presence of obstacles in front of it can be inferred by extracting the waterline. In this work, we present a supervised method for waterline detection, which can be used for low-cost autonomous boats. The method is based on a Fully Convolutional Neural Network for obtaining a pixel-wise image segmentation. Experiments have been carried out on a publicly available data set of images and videos, containing data coming from a challenging scenario where multiple floating obstacles are present (buoys, sailing and motor boats). Quantitative results show the effectiveness of the proposed approach, with 0.97 accuracy at a speed of 9 fps.