Feature Extraction And Recognition Of Inland Waterway Vessels Based On Machine Vision

The introduction of the intelligent traffic management system into inland waterway transport traffic management plays an important role in preventing and controlling the water pollution, protecting the ecological stability and improving the efficiency of inland waterway traffic. Ship-type recognition is an important part of intelligent traffic system.This paper takes the present maritime video monitoring system as a platform,combining the technology of machine vision detection and tracking technology,and then,classify the ship type with the sampling image. Ship type recognition is widely used in some areas,such as, the registration and management of the ship, the investigation of maritime accident, and the channel guide management. On the basis of the feature extracting of the vessels, this paper proposes and designs a set of identification and classification system with multiple classifiers.In this paper, the main content includes the following three parts:1.The interested region extraction. First of all, this paper introduces several kinds of image pretreatment methods: such as scene-adaptive processing, filtering processing,binarization processing, and compares various pretreatment through experiment. Then it adopts a kind of high efficient Region of Interest extraction method by the comparative analysis of several Region of Interest extraction methods and on the base of the practical application of this subject. This method can not only effectively solve the problem of low recognition rate caused by noise,but also reduce the amount of calculation and improve the computing speed.2.The feature extraction. this paper first introduces several common features of an object, such as geometric features, moment features, transformation features and partial features, and then emphatically analyses the advantages of the features of Hu moment and SIFT. Then the general process of the SIFT feature extraction has been introduced in detail. Finally, according to the needs of this topic and the features of the vessels, this paper adopts a method of the ship recognition and classification combined with the global features( Hu features) and the partial features(SIFT features) to improve recognition rate and enhance robustness.3. the ship-type recognition. Firstly, Bag of Words is briefly introduced, then the SVM working principle are expounded in detail, and finally the classification system with multiple features based on Bag of Words is adopted. The system firstly extracts the Hu moment and SIFT features of the ship, then clusters of the SIFT features into features dictionary by K_means algorithm, then corresponding features histogram is generated by adopting the feature words in the dictionary to present the each sample,and then respectively trains the classification by using the SVM.Finally,it designs a decider ruling out the final result. The experimental results indicates: when the parameter is appropriate, recognition rate can reach 91.7%.