Research Of Identification And Tracking Of Moving Underwater Targets Based On Sonar Image

Identification and Tracking of moving underwater targets is an important topic in the field of vision of Autonomous Underwater Vehicle (AUV), and it's the basic of some other subjects such as Target Classification, Analysis of target behavior and so on. It plays a very important role in computer vision process. It's a leading subject which spans many technologies including computer vision, image processing, pattern recognition and artificial intelligence. Therefore, the contents of this paper have broad prospects.The task of identification and tracking of moving underwater targets based on sonar image can be divided into four parts: Acquisition and pre-processing of sonar image, Detection of moving target, Mathematical construction of moving target, Tracking of moving target. This paper researches on the four aspects respectively.Firstly, in the sonar image acquisition and pre-processing module, this paper briefly introduces the Super Seaking DST sonar which is mechanically scanned forward-looking imaging sonar we used, and we also introduce its working mode and its data format. Here we present the pre-processing method; it contains Image Enhancement, image segmentation, mathematical morphology processing and feature extraction.Secondly, we research a variety of moving target detection algorithms. Through study their theory, we choose the most appropriate algorithm we can used on moving target detection based on sonar image, and we test by the experiment that this algorithm is effective in moving target detection based in sonar image.Thirdly, we research mathematical construction of moving target, and it contains two parts mainly: Non-maneuvering target model and maneuvering target model. This paper mainly focuses on the mathematical model of maneuvering target. This part is the mathematical basic for tracking moving target. Finally, we research the moving target tracking technology. This is the core of this paper. We studied the Particle Filter (PF) algorithm, and aiming at the Sample dilution and the Algorithm Complexity Problem of Particle Filter, we proposed the smoothing particle filter based on Particle Swarm Optimization (PSO) and Support Vector Regression (SVR), and it gets good tracking effort in our experiment.