Impact of Surface Reflections in Maritime Obstacle Detection

Abstract

Maritime obstacle detection aims to detect possible obstacles for autonomous driving of unmanned surface vehicles. In the context of maritime obstacle detection, the water surface can act like a mirror on certain circumstances, causing reflections on imagery. Previous works have indicated surface reflections as a source of false positives for object detectors in maritime obstacle detection tasks. In this work, we show that surface reflections indeed adversely affect detector performance. We measure the effect of reflections by testing on two custom datasets, which we make publicly available. The first one contains imagery with reflections, while in the second reflections are inpainted. We show that the reflections reduce mAP by 1.2 to 9.6 points across various detectors. To remove false positives on reflections, we propose a novel filtering approach named Heatmap Based Sliding Filter. We show that the proposed method reduces the total number of false positives by 34.64% while minimally affecting true positives. We also conduct qualitative analysis and show that the proposed method indeed removes false positives on the reflections. The datasets can be found on https://github.com/SamedYalcin/MRAD.

Figures (6)

• Figure 1: (a) Original sample from the dataset; (b) reflections of the boats in (a) are inpainted; (c) dataset image size distribution. Best viewed zoomed in. [Source: (a) - Boats moored on the River Don New Cut by Christine Johnstone, licensed under CC BY-SA 2.0; (b) - Modified from (a) under the same license.]
• Figure 2: (a) Sample false positive due to surface reflection; (b) Inpainted sample showcasing removal of false positives on reflections. Both proposals are from Co-DETR. [Source: Modified from the photo "Boat passing bridge keeper's cottage, Aberchalder by Craig Wallace" under the same license, CC BY-SA 2.0]
• Figure 3: (a) Original sample; (b) inpainted sample. Notice how inpainting enlarges the object downward. Proposals are from Co-DETR. [Source: Modified from the photo by Dimitrios Tzortzis under the same license, CC BY-SA 4.0]
• Figure 4: (a) An example from LaRS dataset. (b) An example from our dataset. Notice the differences is scenery and quality. [Source: (b) - Photo by Glaurent, licensed under CC BY 4.0]
• Figure 5: (a) Proposals on “boat” category; (b) Heatmap of the same category constructed with confidences. [Source: (a) - Modified from the photo "'L'eau-t Cuisine' on the Shropshire Union Canal at Chester by Stephen McKay" under the same license, CC BY-SA 2.0]