WaterScenes: A Multi-Task 4D Radar-Camera Fusion Dataset and Benchmarks for Autonomous Driving on Water Surfaces

TL;DR

Experimental results demonstrate that 4D radar-camera fusion can considerably improve the accuracy and robustness of perception on water surfaces, especially in adverse lighting and weather conditions.

Abstract

Autonomous driving on water surfaces plays an essential role in executing hazardous and time-consuming missions, such as maritime surveillance, survivors rescue, environmental monitoring, hydrography mapping and waste cleaning. This work presents WaterScenes, the first multi-task 4D radar-camera fusion dataset for autonomous driving on water surfaces. Equipped with a 4D radar and a monocular camera, our Unmanned Surface Vehicle (USV) proffers all-weather solutions for discerning object-related information, including color, shape, texture, range, velocity, azimuth, and elevation. Focusing on typical static and dynamic objects on water surfaces, we label the camera images and radar point clouds at pixel-level and point-level, respectively. In addition to basic perception tasks, such as object detection, instance segmentation and semantic segmentation, we also provide annotations for free-space segmentation and waterline segmentation. Leveraging the multi-task and multi-modal data, we conduct benchmark experiments on the uni-modality of radar and camera, as well as the fused modalities. Experimental results demonstrate that 4D radar-camera fusion can considerably improve the accuracy and robustness of perception on water surfaces, especially in adverse lighting and weather conditions. WaterScenes dataset is public on https://waterscenes.github.io.

Figures (9)

• Figure 1: Example scenario from our WaterScenes dataset. For each radar point on the image, the color denotes the range, and the size represents reflected power from the target.
• Figure 2: Samples in WaterScenes. Radar points are projected onto the image plane as colored dots.
• Figure 3: Sensor suite for our USV and coordinate system of each sensor.
• Figure 4: Statistics of objects in WaterScenes. (a) Wide range of object size. (b) Wide distribution of object distance.
• Figure 5: Radar-camera fusion network for the detection benchmark on WaterScenes. Camera images and radar point clouds are fed into the stem layers for feature extraction. Subsequently, the extracted features are processed by the attention mechanism and added along the channel dimension before forwarding into YOLOX-M and YOLOv8-M modules. As a result, the fusion-based network successfully detects boats even when cameras are occluded by waterdrops.