Study On The Restoration And Mosaicing Methods Of Underwater Video

Currently the most advanced underwater operation equipment is Remotely Operated Vehicles(ROVs) or Autonomous Underwater Vehicles (AUVs). In addition to various sensors, underwaterrobots rely heavily on video image information for navigation, underwater object inspection,recognition, and maintenance operations of short range. This is because underwater imagescontain rich color and texture information, which is easy to recognize and understand by operator.However, underwater video images suffer from poor quality due to specific propagation propertiesof the light in the water. Firstly, underwater image is poorly contrasted, blurred, and with verylimited view field because of the light absorption and scattering in underwater environment.Secondly, underwater image usually exhibits non-uniform illumination owing to auxiliary lightingsystem and the motion of underwater platform, thus producing bright spots or dark regions inunderwater image. The above mentioned problems of underwater video make the remote operatorin water surface platform unable to recognize and understand the environment around theunderwater robot, so it is difficult to remote control the robot to accomplish various underwateroperations, thus greatly limiting the practical application of underwater robots. To solve the abovementioned problems, we have investigated the underwater video restoration and mosaicingmethods and have accomplished the following research work.(1) Research of underwater video restoration algorithm. In order to overcome the problemof low visibility and non-uniform illumination for underwater images, an underwater imageformation model with non-uniform illumination is first derived based on the underwater imageformation model with uniform illumination, and then a visibility enhancing algorithm is proposedfrom the developed underwater image formation model. In the algorithm, the mixing image ofmedium scattering light and light change is first estimated and eliminated quickly by using thelow-frequency sub-band information of wavelet transform, then the obtained image is segmentedto bright speck region and scattering region and enhanced according to the segmented resultseparately. The experimental results show that the proposed algorithm can significantly reduce theprocessing time and effectively improve the visibility of underwater degraded images whileeliminating the effect of non-uniform illumination. Moreover, the processing results ofconsecutive frames of video sequence possess good consistency.(2) Research of fast underwater video mosaicing algorithm. The main disadvantages oftraditional underwater video mosaicing algorithm are time-consuming and accumulation ofregistration error. To overcome the problem of registration error accumulation, this paper firstpresents a global consistency based mosaicing strategy which can effectively avoid the erroraccumulation and propagation resulted from the concatenation of pair-wise transformationmatrixes. On the other hand, in order to reduce mosaicing time, the algorithm adopts the keyframe selection and extraction method to reduce the number of mosaicing frames, and uses thecoarse to fine matching method to further decrease the time of feature point extraction andmatching. Moreover, at the stage of panoramic image rendering, the algorithm can adaptivelyselect different rendering methods according to the change of illumination in the overlappingregion to implement seamlessly panoramic image construction. The experimental results show that the proposed algorithm can basically meet the application requirement of underwater video basedlarge area observation.The research could improve the perceptive ability of underwater robot to the environmentaround itself. It is the key technology of underwater observation and operation of short distancefor underwater robots, thereby having significant theoretical research value and promisingapplication foreground.