Abstract
Abstract. Binocular vision system is an essential way for target localization in many fields, which has been widely used as payload of unmanned surface vehicles (USV). High resolution cameras, which can provide richer information, are utilized more often on a USV. This brings challenges of computing tremendous data for target detection and localization in real-time. In this paper, we propose an framework to automatically detect and localize target using high resolution binocular cameras for environment perception of USV. Instead of processing the whole image, the feature extraction and matching are executed within the target region of interest determined by a deep convolution network. Then the target can be localized using triangulation principle with calibrated binocular camera parameters. Experiments show that our proposed strategy can achieve both precise detection and high accurate localization results in real-time applications.
Highlights
Unmanned Surface Vehicles (USVs) are autonomous marine robots which have caused rising attention in recent years
In order to assure the safety of the platform and perform autonomously in complex environment, different types of payloads are mounted on a USV, such as radar, lidar, sonar, camera, GNSS and IMU, etc. (Heidarsson and Sukhatme, 2011; Ji et al, 2014; Liu et al, 2016b; Schuster, 2014; Shi et al, 2019)
We found that especially when the target boat is not far, many extracted and matched points are on the water surface at the top half of the regions of interest (ROIs), which would lead to wrong localization result
Summary
Unmanned Surface Vehicles (USVs) are autonomous marine robots which have caused rising attention in recent years. Many deep CNN networks such as AlexNet (Krizhevsky et al, 2012), VGG (Simonyan and Zisserman, 2014), GoogleNet (Szegedy et al, 2015), etc., have been developed and applied in multiple fields With these deep networks and subsequently proposed concepts such as regularized discarding, residual network, etc., numbers of target detection methods have been proposed such as Fast R-CNN (Girshick, 2015), Faster R-CNN (Ren et al, 2017), R-FCN (Dai et al, 2016), SSD (Liu et al, 2016a), Yolo v1~v3 (Redmon et al, 2016; Redmon and Farhadi, 2017, 2018), etc. The number of the network layers have been significantly increased, the detection accuracy has been improved
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callMore From: The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
