Surface Vehicle Detection and Tracking with Deep Learning and Appearance Feature

Abstract

A real-time vision-based surface vehicle detection and tracking algorithm for the unmanned surface vehicle is proposed in this paper. The algorithm is designed towards the aim of robust close-range vehicle detection and tracking to meet the needs of automatic navigation for the unmanned surface vehicle (USV). We employed a deep convolutional network to obtain high-performance object detection. Then a data association-based multiple target tracking method is implemented by combining the appearance feature and the estimation of the motion state through the Kalman filter. The experimental results show that compared to SORT tracking algorithm, the proposed method can reduce the number of identity switches and better complete the detection and tracking of surface vehicles.