Interference intention classification of moving obstacles used for USV collision avoidance

Abstract

Unmanned Surface Vessels (USV) may interfere with ships in the same lane during navigation and may be affected by ships that actively collide or intercept the USV. Evidently, the latter poses a greater threat than the former. This paper proposes the concept of active and inactive interference. It is necessary for the USV to accurately classify the vessels’ intentions to conduct a suitable collision avoidance strategy. However, the existing research on collision avoidance of USV focuses on avoiding moving obstacles and rarely considers the interference intention of dynamic obstacles for USV when choosing collision avoidance strategies. This paper proposes an algorithm to recognize the interference intention of ships by combining visual classification with a Support Vector Machine (SVM). First, a Convolutional Neural Network (CNN) is proposed to distinguish the merchant ship from a high-performance ship that actively interferes with the USV with high probability. A high-dimensional feature dataset was designed to illustrate the navigation characteristics of an obstacle, and Principal Component Analysis (PCA) was then employed to reduce the dimensionality of the dataset for obstacle classification. A modified SVM is presented to classify obstacles into active and inactive interference intention categories. Moreover, an escape algorithm based on an improved artificial potential field method is proposed for vessels with active interference. The simulation and experimental results clearly show that the USV can successfully identify the interference intention of obstacles and adopt a specific collision avoidance strategy for obstacles with active interference intentions. Using this algorithm, the USV can safely avoid obstacles with different interference intentions.

Keywords

• Unmanned Surface Vessels (USV)
• Interference intention classification
• Machine vision
• Collision avoidance
• Machine learning