Ranging And Recognition Of Water Targets For USV Based On Binocular Stereo Vision

In recent years,with the gradual depletion of the land resources and the deterioration of the ecological environment,countries around the world have begun to turn their attention to the vast ocean territory.The development of marine equipment is particularly important in this battle for marine resources,and the Unmanned Surface Vehicle is one of the emphases for research.This paper mainly studies the environment perception technology of the USV.Based on the analysis of the advantages and disadvantages of common sensors,a ranging and recognition system based on binocular stereo vision for the USV is proposed.Firstly,this article introduces the basic concepts of binocular stereo vision ranging such as camera coordinate conversion,camera distortion and stereo matching.Aiming at camera calibration technology,this paper uses the Zhang Zhengyou calibration method to calibrate the camera,and uses the MATLAB camera calibration toolbox to complete the calibration experiment of the binocular camera.In stereo matching,in view of the shortcomings of the semi-global stereo matching algorithm,this paper proposes a disparity range estimation method based on FAST,and improves the matching cost computation by using the fusion of gradient,color,Census Transform and disparity range.Experiments show that the improved algorithm model improves the overall disparity accuracy effectively and meets the requirements for ranging of water targets basically.Secondly,in the aspect of target detection,YOLOv3 is selected as the research scheme of this paper.This article divides the common targets of water navigation into five categories:ship,sailboat,reef,glacier,and aids to navigation.The 2,000 images collected are manually labeled,and a data set is made according to the PASCAL VOC format.On the basis of YOLOv3,GIoU is used to optimize the loss function,and data augmentation methods are used to increase the amount of training data and improve the generalization ability of the model.The experimental results prove that the mAP of the trained model is improved by3.19%,and it can adapt to various practical scenes.Finally,this article expands the system further.While exploring the multi-target tracking principle of the Tracking-by-Detection method,the process of the SORT algorithm is studied deeply.A target tracking algorithm combining YOLOv3 and SORT is written to achieve real-time tracking of targets.A maritime target ranging and recognition system based on thePyQt5 framework is designed.Experiments show that the system basically meets the requirements of real-time and accuracy,which has certain guiding significance.