Research On Detection And Tracking Algorithm For Ship In Inland Waterway

The inland maritime supervision plays an important role in the development of economic zone of Yangtze River,the river freight volume of which rank first in the world.Traditional methods of maritime supervision,including camera monitoring and artificial cruise,are unable to meet the demands of the inland maritime supervision nowadays.While the intelligentized supervision system can effectively attribute to a greater efficiency which is lacking in the traditional ways.Detection and tracking algorithm of inland ships,the related researches of which have been put in practice such as inship traffic statistics,accident warning and detection of violations increasingly,is the important part of the intelligentized maritime supervision system.This thesis aims at solving the problems encountering in the process of ship detection and tracking,including the interference of water surface on detection algorithm and the influence of large target,similar color and occlusion.The video data of inland waterway collected by webdome has been stored in the form of data stream down in the local hard disk.And the off-line data will be transmitted to the computer periodically and be used for processing analysis later.The main works are as follows:(1)Image sequence preprocessing and management of trajectories.First,the videos will be converted from YV12 to RGB24 which enable pixels to be processed in RGB color space.Reduction of the memory footprint and calculation work by down-sampling on the original images with different sampling coefficient will be put up.The most suitable sampling coefficient which could balance the speed and effect of algorithm will be determined by experiments.The system is established based on the rules of managing trajectories which is designed with the result of detection and tracking.(2)Research on ship detection algorithm in inland river.In order to solve the problem that the foreground object will be “eaten-up”,this thesis proposes a new method to measure the background complexity based on region information.The background complexity is used to control the background updating rates in the detection algorithm which make the result complete.Meanwhile detecting the salient region of images thus the thresholding result could be used to fill the hole of foreground.According to the characteristics of the ghost that it is easy to be taken as background because of its similarities to the background,a scanning window will be used to check the result by eliminating ghosts.If the pixel value of the point is closed to the background around it,that point could be judged as a ghost point.(3)Study the methods of ship tracking.First,the shortcomings of the classical tracking algorithm under occlusion have been analyzed.Then,a novel sparse optical flow tracking algorithm based on the fail feature point detection is proposed too,based on the characteristic that shipsare artificial rigid body is rich with corners.The Shi-Tomasi corner detection algorithm is used to detect the feature points of the ship,and the LK algorithm is used to calculate the optical flow of the feature points in the image sequence.Fail points will be detected based on the optical flow's size,direction and the matching degree of template in the process of tracking.Finally,the target position is located based on the tracking result of the effective points.The experiments are carried out to verify the effectiveness of the ship detection and tracking algorithm.And the detection results show that the best recall of 0.93,F1 measure of 0.82.The tracking results show that the best root mean square error is about 1.8 pixel distance.