Research Of Ship Tracking And Type Recognition Based On Computer Vision

At present,with the rapid growth in the number of motor vehicles,more traffic accidents and congestion have also taken place in the land,while inland waterways are also facing the same pressure.It is of great significance for the effective control of water pollution,the protection of river ecological stability and the safety of waterway transportation by drawing on the experience of land traffic management and introducing intelligent traffic management system into the management of inland waterway transport.Ship identification is an important part of shipping management.In this paper,the marine video surveillance of Zhongshan City is taken as the research background and based on the "Marine Command Platform" already established in the city,the identification of ship types carried out in the actual need of inland river area.Based on the theory and method of computer vision,this paper extracts and traces the ships in the surveillance video,and then identifies the ships based on the extracted features.The research contents include the following aspects:(1)The extraction of ship motion area.Firstly,image preprocessing is introduced,including gray-scale,image enhancement and binarization processing,and the pretreatment is analyzed and compared by experiments.Then,the experimental results and advantages and disadvantages of common methods are analyzed in the extraction of motion areas.(2)Target tracking.Under the premise of extracting the target region,Camshift algorithm based on the histogram of color feature has the characteristics of small computation and high real-time performance.Ships are easily lost due to obstruction.So Kalman algorithm is used for target prediction to realize continuous tracking and prevent target loss.(3)Feature extraction.The commonly used features are introduced,such as geometric features,moment features,histogram statistical features,corner features and so on,and extract the eigenvalues.By analyzing the data,the ship is classified by inputting ratio feature,the moment feature and the angle ratio feature.(4)Ship identification.This paper adopts support vector machine to classify ships,by comparing the single feature,multi-feature and all the characteristics of the experiment,select the correct combination of high rates.And then use the idea of cross-validation to improve the testing process and training process for the accuracy.Finally,the paper chooses the background difference method of single Gaussian modeling to extract the motion region.The Camshift and Kalman filter are used to track the target.Also it can track the target effectively.Then the proportionality,moment characteristic and angle feature are selected as the classification of ship classification.Finally,use the idea of cross-validation to improve the testing and training process to prove the idea of the feasibility and correctness,which achieve a high accuracy.