Research On Classification Method Of Moving Target At Sea Based On MPEG-7 Standard

Target classification is an important research content in the field of intelligent video surveillance, which is the basis of analysis and understanding of the target’s behavior in video. In general, the specific target classification method is based on the features of target to be classified, there is no classification method suitable for all types of target. At present, a large number of studies have focused on the classification of land targets, but the special study on the classification of sea targets is relatively less. Automatic classification of sea video targets can improve the perception of crew for the environment around the ship, but also can be widely used in maritime security, maritime management, channel management, maritime search and rescue and other fields.This paper introduces MPEG-7 standard to the application of ship target classification in maritime video, which can be used to standardized description of all kinds of multimedia information. The paper presents a classification method of moving target at sea based on MEG-7 standard. Firstly, through analyzing visual features of ship targets at sea, we found that shape features have obvious separability among various types of ships, this paper makes use of shape feature to carry on classification of ship targets.Secondly, considering translation, scaling and rotation of ship targets in video, meanwhile targets can’t be detected accurately and completely, Hu invariant moments and ART coefficients are selected to describe the region shape features of ship targets. According to the comparative analysis of feature value of Hu invariant moments and ART coefficients, this paper selects seven feature values which have best separability to form feature vector.Thirdly, in view of the fact that there is no uniform standard format for storage of sample feature files, which is often not conducive to reuse of existing feature library. According to the uniform format of MPEG-7 standard, the feature description file is generated by the vector of ship target regional shape feature, and the shape feature library for ship targets is established based on the model of various ships.Finally, the classification experiments of five kinds of ship model images and detected ship targets in the sea video are carried out by using KNN template matching and BP artificial neural network. The results show that a higher correct recognition rate has been achieved using two classification methods in respect of ship targets classification.