Research On Underwater Multi-target Detection Technique Based On Forward-looking Sonar

Underwater targets detection and feature extraction techniques were important parts ofunderwater vehicles for underwater operations, the sonar system served as one of visionsystems to provide information for operations and obstacle avoidance of underwatervehicles. Sonar images were quite different from optical images, so did the imageprocessing methods.This project came from the Natural Science Foundation of China “UnderwaterMulti-target Locating Method Research Based on Visual and Ultrasonic ProximateSensors”, the contents of this paper were important for it. With the underwater operationtest platform constructed by remotely operated vehicle “OUTLAND1000”which fittedout a single beam mechanical scanning forward-looking sonar as the research object, thispaper made multi-target detection based on acoustic visual system.Firstly, communicational protocol of the forward-looking sonar was analyzed, and thecommunication of personal computer and sonar system was realized by converting RS485to RS232serial port, sonar echo data were acquired by programming to make secondarydevelopment on software. Secondly, using coordinate transform method and beaminterpolating method to generate sonar images, using piecewise linear grayscaletransformation, median filtering algorithm, threshold segmentation which adopted anadvanced converting two-dimensional to one-dimensional Otsu adaptive thresholdsegmentation method for shortening the calculation time and getting better results, andmorphology closing algorithm to make image preprocessing. Thirdly, edge detection andedge thinning method, or contour tracking method were used to acquire edges with singlepixel of multi-target. Lastly, an advanced chord midpoint Hough transform ellipsedetection method was proposed based upon sonar characteristics and geometriccharacteristics of underwater targets to realize feature extraction of elliptic shape which could tell the number and location of underwater targets. The experiments showed that theabove methods worked well for multi-target feature extraction, and provided usefulinformation for operations and obstacle avoidance of underwater vehicles.