A Study On Recognition Of Ship Anchor Based On Video Image Analysis

In our country,ship mooring in inland river usually rely on anchor chain or cable fixed in anchorage.However,sediment loosening and the broken cable cause anchor accident.Serious economic losses has caused by the collision of the ship and the bridge caused by the anchor.In recent years,with the development of water transportation,the number of ships and the average weight of the overall growth,the loss of a single anchor accident is increasing.Lacking of effective means of supervision,there did not provide safety monitoring service in anchorage management.Conventional walking anchor recognition method is focused on mechanical model based on the combination of sensor detection,GPS positioning and radar ranging and so on.The mechanical model is complex and the calculation of cable is difficult to obtain overall.There is obvious mistakes based on GPS positioning sometimes,what's worse,the ship is out of power while mooring normally,which lead the positioning signal is out of access.So far,ranging based on radar only provide the target point information,without color,shape,size and other relevant information,which make it difficult to make precise judgments.Analysis based on video image comes to mature and video data is rich in information,good real-time performance.In the vehicle detection and tracking recognition have much achieve.However,there are obvious differences in the water target movement environment and the road environment.This thesis did some research on the difficulties in the recognition of the ship's anchor through video image analysis method.The main work and results are as follows:1 Method of ship detection in anchorage water environment is studied,a method of maximum background distribution to improve GMM is proposed.Firstly,the characteristic of the surface wave is analyzed,and the difference between the water background and the road background is compared.Secondly the background is described by the Gauss distribution of the maximum number,which can overcome the noise caused by the false detection of the surface movement.Lastly according to the surface reflection target obviously,put forward the method of eliminating reflection mirror characteristics of ship.2 Method of tracking about non-navigation ship's motion is studied,according to dragging motion characteristics,and Kalman filtering is pulled-in to improve the estimation accuracy.The classical MS is not accurate in the fast target tracking,and combined with the initial position by Kalman to guide MS vector offset and improve the accuracy.In the color image,according to the calculation of the matching of the rectangular frame,the center position is calculated.3 An incremental updating method based on distance is proposed to expand the training set.Two parameters of the classical DBSCAN algorithm are fixed,which makes the clustering results are sensitive to the parameters.Analysis in ship anchor activity distribution uneven characteristics.KNN matrix was used to get multiple neighborhood radius,so as to ascendingly cluster with normal anchoring position data.The incremental updating method is used to try to cluster test data to normal cluster,and the training set is extended to overcome the defect of long duration of data acquisition and reduce the false detection rate.4 The simulation experiment was carried out to verify methods proposed in this thesis and results were analyzed.Data gained in detection and tracking of tanked ship were used to dynamically cluster analysis.