Study On Visual Object Tracking Methods Of Homeward-Bound Unmanned Surface Vehicle

As a new surface warfare platform,USV plays an important role in both civil and military fields,and it has broad development prospect.The visual system based on visible light is more abundant in color and texture compared with radar or infrared target detection.Target template matching,visual object identification and visual object tracking are important technologies when USV completes the task and tracks the known static port target.Based on a large number of experiments,this paper focuses on visual tracking which has the known static port target.Image preprocessing of visible light,identification and tracking of visual target are mainly realized in this paper.The main contents are as follows:(1)The correlation processing of visible light images,the target characteristics and the characteristics of noise during the homeward-bound voyage are analysed in this paper.To realize the pretreatment of visible light of water surface,denoising method and highlight removing are also analysed based on algorithms and experiments.(2)In this paper,the identification of the known port target is realized by the method of matching the feature of the target template library and the scene feature.To ensure the accuracy and real time of USV's target identification,some methods of keypoint detecting are introduced.In this paper,the original algorithm of SURF is improved by combining SURF feature extraction and LBP feature description.The algorithm of euclidean distance and RANSAC are used to complete the image matching.Based on a large number of experiments,the improved matching algorithm and the commonly used matching algorithm based on local invariant feature are compared,and the target recognition based on the improved matching algorithm is realized in this paper.(3)The correlation filter template feature method and color feature method are combined by STAPLE.It is helpful to solve the problems such as continuous change of the scale,rapid target movement and light change in the course of the USV,but it is not good in the motion blur scene caused by the wave and the mechanical vibration of USV.In this paper,the background information is reasonably increased when the filter is trained,so that the algorithm has strong adaptability to motion blur.(4)The algorithm is optimized in this paper in order to meet the real-time requirement of the algorithm.The speed of the algorithm is improved by PCA reduction and QR decomposition.(5)Target will lose because of the error accumulation when target is blocked.In this paper,the tracking failure due to occlusion are effectively solved by combining the tracking of keypoint matching and background-aware STAPLE.