Learning Maritime Obstacle Detection from Weak Annotations by Scaffolding

Abstract

Coastal water autonomous boats rely on robust perception methods for obstacle detection and timely collision avoidance. The current state-of-the-art is based on deep segmentation networks trained on large datasets. Per-pixel ground truth labeling of such datasets, however, is laborintensive and expensive. We observe that far less information is required for practical obstacle avoidance -the location of water edge on static obstacles like shore and approximate location and bounds of dynamic obstacles in the water is sufficient to plan a reaction. We propose a new scaffolding learning regime (SLR) that allows training obstacle detection segmentation networks only from such weak annotations, thus significantly reducing the cost of ground-truth labeling. Experiments show that maritime obstacle segmentation networks trained using SLR substantially outperform the same networks trained with dense ground truth labels. Thus accuracy is not sacrificed for labelling simplicity but is in fact improved, which is a remarkable result.

Online Estimation of Ship Dimensions by Combining Images with AIS Reports

“…Bovcon et al [14] developed a deep encoder-decoder framework (the water obstacle separation and refinement network) for autonomous crewless ship navigation that extracted the contours of several ship targets. Ust et al [15] introduced a scaffolding learning regime (SLR) that trained an obstacle detection segmentation network under weak supervision for individual ship contour extraction. Kelm et al [16] trained a CNN to identify central pixels; the network recognized a part of an input image and calculated a rotation angle, and used the central pixel to describe the upcoming directional change in the contour.…”