A Multi-modality Sensor System for Unmanned Surface Vehicle
- PDF / 1,599,501 Bytes
- 16 Pages / 439.37 x 666.142 pts Page_size
- 49 Downloads / 180 Views
A Multi‑modality Sensor System for Unmanned Surface Vehicle Hao Liu1,2 · Jie Nie1 · Yingjian Liu1 · Yingying Wu1 · Hanxing Wang1 · Fangchao Qu1 · Wei Liu1 · Yangyang Li1
© Springer Science+Business Media, LLC, part of Springer Nature 2019
Abstract The onboard multi-modality sensors significantly expand perception ability of Unmanned Surface Vehicle (USV). This paper aims to fully utilize various onboard sensors and enhance USV’s object detection performance. We solve several unique challenges for application of USV multi-modality sensor system in the complex maritime environment. By utilizing deep learning networks, we achieved accurate object detection on water surface. We firstly propose a multi-modality sensor calibration method. The network fuses RGB images with multiple point clouds from various sensors. The well-calibrated image and point cloud are input to our deep object detection network, and conduct 3D detection through proposal generation network and object detection network. Meanwhile, we made a series of improvements to the system framework, which accelerate the detection procedures. We collected two datasets from the real-world offshore field and the simulation scenes respectively. The experiments on both datasets showed valid calibration results. On this basis, our object detection network achieves better accuracy than other methods. The performance of the proposed multi-modality sensor system meets the application requirement of our prototype USV platform. Keywords Multi-modality sensor · Unmanned Surface Vehicle · Object detection · Sensor calibration
1 Introduction Unmanned Surface Vehicle (USV) plays an important role in the scientific investigation, marine monitoring and disaster assistance [1]. Modern USVs usually equipped with various sensors such as camera, radar, positioning sensors. These sensors can obtain diverse data form such as image, depth, pose, position. In many survey * Jie Nie [email protected] 1
Department of Computer Science and Technology, Ocean University of China, Qingdao 266100, Shandong, China
2
National Laboratory for Marine Science and Technology (Qingdao), Qingdao 266000, Shandong, China
13
Vol.:(0123456789)
H. Liu et al.
activities, the USV already has the capability of autopilot [2]. However, most USVs use only elementary sensors and algorithms to implement collision avoidance. They do not take full advantage of the multi-modality sensor [3]. Consider Unmanned Ground Vehicle’s (UGV) autonomous driving, fully utilizing multi-modality sensors provides strong support for object detection and further leads to advanced autonomous operation for missions. However, Compared to other unmanned platforms such as Unmanned Aerial Vehicle (UAV) and UGV, USV faces some unique challenges in processing data obtained from multimodality sensors. The operating conditions of USV are usually abominable. Temperature, vibration, and illumination will impact the sensors. This disturbs algorithms. In this paper, we proposed a multi-modality sensor system for USV and realized
Data Loading...
