Tracking Algorithm Of Inland River Ship Based On Correlation Filter

Recently,with the rapid economic development,convenient and fast water transport,has become an important mode of transport.However,some shipping participants have a weak shipping awareness,and some behaviors such as retrograde,overloading and overrunning also bring safety risks to water transportation.In addition,the water transport environment is poor,especially the domestic inland river basin,some waters are shallow,curved,navigation environment is poor,easy to cause traffic accidents.Therefore,it is a major and important project to effectively control the navigation environment,record the ship’s route,monitor the ship’s operation behavior,and timely give early warning to the illegal ships and the complicated and dangerous navigation environment.Before judging the behavior of the inland river ship,it is necessary to accurately locate the target ship and keep track of the target ship during the movement.Ship tracking technology can extract ship features and obtain ship details(location and size,etc.).Therefore,it is one of the key technologies to realize accurate tracking of inland river ships.This paper systematically analyzes the attributes of video sequences in inland rivers.Around the "how to accurately and in real time track the ship in the river,to overcome the ship in the process of occlusion interference,scale variation" this key problem.In order to verify the effectiveness and robustness of the algorithm,the improved algorithm is proposed and the videos of inland river ships in different scenarios are tested.Specific completion of the following aspects of work.In order to verify the effectiveness of the algorithm,this topic manually annotated the ship data set and established the public data set according to the classification of 10 attributes.In this way,the tracking accuracy and success rate of the algorithm under different attribute videos and different length sequences are verified,which is also one of the contributions of this project.In order to achieve accurate tracking ships in the research target,this topic first puts forward an improved algorithm based on correlation filter,the introduction of context clues,combined with the time continuity and spatial correlation,and proposes the optimization of regression method,can more accurately fitting the space structure of the target,to improve the algorithm of tracking accuracy and the success rate of ship’s condition.The feature fusion method is further introduced to extract CNN features by using VGG-M network and combine with Hog and CN features to solve the scale variation of ships.Based on the data established above,experimental test results show that,compared with some advanced algorithms,the improved method with feature fusion based on correlation filter have better tracking accuracy and success rate in the case of occlusion interference and scale variation.Under the OPE evaluation,the occlusion tracking success reaches 0.774,and the scale tracking success reaches 0.832.