Research Of Detection Method For Man-made Objects In Underwater Images

Currently the most advanced underwater operation equipment is RemotelyOperated Vehicles (ROVs) and Autonomous Underwater Vehicles (AUVs).Underwater robots should first detect the presence of the underwater targets toimplement the right navigation. However, man-made objects detection in complexunderwater scene has two problems. Firstly, underwater image is blurred and colorfaded because of the light absorption and scattering in underwater environment, thusthe color and texture features cannot be the detection features of underwaterman-made objects. Secondly, underwater scene is complex, since it contains a varietyof plants, rocks except man-made objects, so that the traditional thresholdsegmentation method cannot be used. Therefore, the real-time underwater man-madeobjects detection in complex underwater scene is an urgent problem in underwaterrobot applications. To solve the above mentioned problems, we have investigated theunderwater man-made objects preprocessing and detection methods and haveaccomplished the following research work.(1) Research of underwater image preprocessing algorithm. Due to the specialunderwater environment, underwater light has scattering and attenuation effect, andunderwater video image is poorly contrasted and blurred. Therefore, we should firstdo some preprocessing to the underwater image. This dissertation proposes acombination of edge detection algorithm based on gradient histogram and iterationmethod according to the characteristics of underwater image. Firstly, we use a medianfilter to reduce the effects of noise, and then calculate the gradient magnitude usingSobel operator, and perform non-maxima suppression to gradient magnitude. Finally,gradient histogram and iteration method is used to determine the adaptive edgedetection segmentation threshold value for the binary image. The experimental resultsshow that the proposed algorithm can get good edge detection results, thus beingsuitable for further image segmentation.(2) Research of underwater man-made objects detection algorithm based on linefeature. To overcome the time consuming problem of man-made objects detection incomplex underwater scene, this dissertation selects straight line features to detectunderwater man-made objects. Man-made objects are generally constituted by straight lines according to observe a large number of underwater videos, and the line feature isnot affected by illumination changes and scattering attenuation. Therefore, we candetermine the presence of a candidate man-made object whenever there are some longstraight lines. In order to fast detect man-made objects, the appropriate detection scaleis first determined by using Daub5/3wavelet transform, and then edge detection isperformed on the determined low frequency subband. Finally, the line features ofman-made objects are detected by employing a refined Hough transform basedmethod. The algorithm is performed on the determined low frequency subband image.Since only the salient edge points are detected, this algorithm has high real-timeperformance. The experimental results show that the proposed algorithm canaccurately find out the straight lines existed on man-made objects in complexunderwater background and has favorable real-time performance, thereby satisfyingthe practical application requirement of underwater video based on man-made objectsdetection.