Ship Recognition And Tracking Algorithm Based On HOG Features

In the waterway,the type of ships is increasingly diversified,and the ship sailing trajectory is becoming more and more complicated.Monitoring of ship activities requires real-time-tracking ability,as well as the recognation of ship targets.The traditional video-object tracking algorithms have the flaws of large computational cost,limited tracking accuracy and lack of recognition ability,it causes that it is difficult to dispatch ships..Therefore,it is necessary to design a real-time,accurate ship target tracking algorithm with recognition ability.The research of this paper includes three aspects:feature extraction of ship target,recognition of ship targets,and tracking of ship targets.In the aspect of target feature extraction,algorithem of this paper chooses the HOG feature of ship target to reduce the influence of other noises on ship recognation in the water environment by comparing the advantages and.disadvantages of various image features.In the aspect of recognition of ship targets,the SVM model has certain classification and recognition ability to the original HOG features of the target.However,not all the target's HOG featuresare valid.Some of them are redundant,some of them are noise.And the complexity of the traditional model is too high.Therefore,the algorithm of this thesistakes the advantages ofsequence forward selection method to remove noise and redundant features.However,because the cross-validation method is adopted when dealing with the training sample data set,and the sequence forward selection method can only add features but not remove features,so the best features selected by the original algorithmare uncertain.In view of the above-mentioned defects,based on the sequence forward selection method,one approach of this thesis includes the construction of proposed feature scoring system to select the best features.In the aspect of target tracking,algorithm of this thesis is based on STC tracking algorithm to determine the target's location.STC algorithm performs well on tracking speed and reliability.However,when ship target is covered by other objects,the performance of STC algorithm is not expected.One approach of this thesis modified this flaw of STC algorthim based on template-matching detection.And the improved STC algorithm performs better than the original.Experiments showed that the ship recognation and tracking algorithm can track the ship target in real time,stable and accurate.