| The rich Marine resources are the treasure land endowed by nature to human beings.Since the 21 st century,Marine resources have gradually become the contested land for the development and utilization of all countries in the world.Due to its excellent performance in deep-sea operations,Unmanned Underwater Vehicle(UUV)plays an important role in topographic and geomorphic detection,life-saving search,fishing and other aspects.Due to its high cost and limited energy carried by itself,UUV cannot be a one-time operating tool.Therefore,after UUV is completed,it needs to be recovered for energy supply and information collection,and then work again.In this paper,based on UUV short-range optical vision recovery,the multi-target tracking in the recovery process is deeply studied.The specific research contents are as follows:Firstly,the monocular vision system in UUV recovery process is introduced.The UUV fork-column recovery strategy is designed.The hardware parameters required by the visual system are described and the array of asymmetric target guidance light source is designed.The system model and its corresponding coordinate transformation calculation formula are discussed.The localization method of UUV four degrees of freedom is given.Secondly,the image processing and light source recognition in UUV recovery process are introduced.Images acquired by UUV through its own underwater camera need to be preprocessed to highlight the features of the light source and then extract relevant information.Considering that underwater noise is mostly randomly distributed,the selection of median filtering can well suppress noise and ensure image details.Moreover,the underwater environment is mostly dark color,and the gray level of the target light source and the background is greatly different.An adaptive threshold segmentation method is designed to effectively segment the foreground and background information of the target light source,and the shape of the light source can be modified by smoothing the edge through morphological operation.Then,a method of extracting the connected domain features of target light source based on Blob analysis is introduced,and a light source recognition method based on geometric rigid body location is designed.An occlusion discriminator is constructed by Blob descriptor.In the case that the target light source is occluded,a fitting modification based on edge detection and Hough circle detection is designed.Thirdly,the tracking method of target guidance light source in UUV recovery process is introduced.Firstly,the basic theory and derivation process of Kalman filter are introduced,the linear Kalman filter is extended to the extended Kalman filter to predict the position information of the target light source,and the disturbance of the observer is corrected by using partial Kalman filter when the target light source is occluded.Then the traditional Hungarian algorithm is introduced,and the algorithm is improved.The optimization cost matrix is designed by combining the Euclidean and distance,light source area,Io U and other features,and the detection target and historical trajectory are correctly matched.The data association method involved is also applicable to the case of information loss of the target light source.Finally,in order to verify the engineering effectiveness of the tracking algorithm proposed in this paper,a pool simulation experiment is designed.Firstly,the experimental environment and hardware composition are introduced.Then,a target tracking scheme is designed to simulate the 4-DOF variation of UUV in the recovery and tracking process,including the change of vertical distance,horizontal position and heading Angle.The tracking scheme of feature information loss when the target light source is out of sight and under occlusion is designed.Finally,the tracking simulation experiment is carried out,and CLEAR MOT multi-target tracking evaluation index is used to evaluate the algorithm performance. |
