| Title |
Former Unmanned Surface Vehicle Detection Based on Improved Convolutional Neural Network |
| Authors |
(Bangqian Ao) ; 김동헌(Dong Hun Kim) |
| DOI |
https://doi.org/10.5370/KIEE.2021.70.10.1488 |
| Keywords |
Object detection; Accuracy; Tracking system; Monocular camera |
| Abstract |
This paper proposes an approach to the real-time implementation of a convolutional neural network (CNN)-based object detector for a former Unmanned Surface Vehicle (USV). The original network VGG-16 of the Single Shot MultiBox Detector (SSD) is first replaced with ResNet-18, as the basic feature extraction network. The classifying network is then redesigned by reducing half of the convolutional kernel numbers, where kernel sizes of 1×1 and 3×3 are mainly used. Simultaneously, a monocular camera installed on the tracking system, is used to calculate the distance and azimuth of the former USV. The experimental results show that the proposed method has advantages of higher accuracy and lower computational complexity, compared with other existing approaches.
Therefore, the proposed approach can be efficiently used on real-time tracking systems. |