Research On Tracking System Of Forward-looking Sonar Image

Many resources are buried in the ocean, so People pay more and more attention to for thedevelopment of ocean. The current detection method of Marine include magnetic detection,sound detection, light detection and infrared detection, and so on. But the propagationdistance of sound waves in the water is more than others, cause the attenuation of soundwaves in water attenuation is smaller. Sonar is the main equipment of acoustic detection ofocean.Forward-looking sonar is a kind of active sonar which is installed in the front of the shipor underwater robot. Forward-looking sonar can not only detect the condition in the sea, butalso can to target positioning, determine size and shape of the target information. Especiallyfor some underwater robot, forward-looking sonar is equivalent to the eyes. They can help theunderwater robot to track the target. So the study of forward looking sonar image targettracking system has very realistic significance.Acoustic detection and tracking system of underwater target is the focus of theunderwater research field. It is also an important component of underwater robot which is asystem of intelligence. It contains image processing and target tracking knowledge. It haspractical significance to promote ocean development. Therefore, it appears to be important toresearch on this.This system has four part: image pre-processing, matching of target, feature extraction,target detection.Firstly, Loading and showing program the SeaBat7128sonar data is carried out in theVs environment. After analyzing ordinary image interpretation, a quadratic interpretationbased on back mapping is created. The experiments prove its good property.Secondly, considering the truth that the image of sonar and the constantly changingbackground，the method of motional object detecting which combines clustering-based imagesegmentation with variable frame has been proposed.Thirdly, comparing the difference between optical image and the sonar image in thefield of feature extraction, a method of feature extraction combined moment invariants andwavelet transform was proposed after analyzing the characteristic of the forward-lookingsonar image. This is the foundation of the next work which is target tracking.Finally, based on the fact that the shape of the target in the sonar image was changing inevery frame, the method combined particle filter and deformable template was proposed andproved by the experiment. The system was designed in the Vs2008environment using C++and Opencv.