Ship collision avoidance methods : State-of-the-art

Recognizing objects from camera images is an important field for researching smart ships and intelligent navigation. In sea transportation, navigation marks indicating the features of navigational environments (e.g. channels, special areas, wrecks, etc.) are focused in this paper. A fine-grained classification model named RMA (ResNet-Multiscale-Attention) based on deep learning is proposed to analyse the subtle and local differences among navigation mark types for the recognition of navigation marks. In the RMA model, an attention mechanism based on the fusion of feature maps with three scales is proposed to locate attention regions and capture discriminative characters that are important to distinguish the slight differences among similar navigation marks. Experimental results on a dataset with 10260 navigation mark images showed that the RMA has an accuracy about 96% to classify 42 types of navigation marks, and the RMA is better than ResNet-50 model with which the accuracy is about 94%. The visualization analyses showed that the RMA model can extract the attention regions and the characters of navigation marks.

/pdf/visual-recognition-based-on-deep-learning-for-navigation-tzh6rzwhge.pdf

Visual Recognition Based on Deep Learning for Navigation Mark Classification

A ship collision avoidance system for human-machine cooperation during collision avoidance

Avoiding collisions with other objects is one of the most basic safety tasks undertaken in the operation of floating vehicles. Addressing this challenge is essential, especially during unmanned vehicle navigation processes in autonomous missions. This paper provides an empirical analysis of the surface target detection possibilities in a water environment, which can be used for the future development of tracking and anti-collision systems for autonomous surface vehicles (ASV). The research focuses on identifying the detection ranges and the field of view for various surface targets. Typical objects that could be met in the water environment were analyzed, including a boat and floating objects. This study describes the challenges of implementing automotive radar sensors for anti-collision tasks in a water environment from the perspective of target detection with the application for small ASV performing tasks on the lake.

/pdf/the-empirical-application-of-automotive-3d-radar-sensor-for-35zylu9mtj.pdf

The Empirical Application of Automotive 3D Radar Sensor for Target Detection for an Autonomous Surface Vehicle’s Navigation

With the development of marine technology, the research of unmanned surface vehicles (USVs) has received wide attention from the world. Designing effective collision avoidance algorithm in dynamic sea environment is necessary for the navigation of USVs. This paper describes a collision avoidance system (CAS) with COLREGs compliance to improve autonomous navigational ability of USV. For small podded propulsion USV, a three-DOFs maneuvering model is established and the parameters of model are determined by the parameter identification method based on field experiments. About the COLREGs rules, the different maneuvering behaviors are summed up according to encounter situation and collision risk assessment. In the CAS, finite control set model predictive control is adopted, and COLREGs rules are taken as an evaluation standard in the quality function. Finally, the system is verified to avoid multiple dynamic obstacles and compliance vessel in the simulation and real environment, and the results of collision avoidance under four encounter situations are described in detail.

Collision Avoidance of Podded Propulsion Unmanned Surface Vehicle With COLREGs Compliance and Its Modeling and Identification

Unmanned surface vehicles (USVs) have become a focus of research because of their extensive applications. To ensure safety and reliability and to perform complex tasks autonomously, USVs are required to possess accurate perception of the environment and effective collision avoidance capabilities. To achieve these, investigation into realtime marine radar target detection and autonomous collision avoidance technologies is required, aiming at solving the problems of noise jamming, uneven brightness, target loss, and blind areas in marine radar images. These technologies should also satisfy the requirements of real-time and reliability related to high navigation speeds of USVs. Therefore, this study developed an embedded collision avoidance system based on the marine radar, investigated a highly real-time target detection method which contains adaptive smoothing algorithm and robust segmentation algorithm, developed a stable and reliable dynamic local environment model to ensure the safety of USV navigation, and constructed a collision avoidance algorithm based on velocity obstacle (V-obstacle) which adjusts the USV’s heading and speed in real-time. Sea trials results in multi-obstacle avoidance firstly demonstrate the effectiveness and efficiency of the proposed avoidance system, and then verify its great adaptability and relative stability when a USV sailing in a real and complex marine environment. The obtained results will improve the intelligent level of USV and guarantee the safety of USV independent sailing.

Radar-based collision avoidance for unmanned surface vehicles

/pdf/design-of-the-motion-control-system-for-a-long-endurance-auv-3hlnec3gor.pdf

Shi-qi Zhao

Papers

Radar-based collision avoidance for unmanned surface vehicles

Design of the Motion Control System for a Long Endurance AUV