Research On The Technologies Of Underwater Image Segmentation And Object Location Based On Monocular Vision

Autonomous underwater vehicles is an important equipment for ocean carrier. It shouldobtain a variety of underwater environmental information to complete the mission, especiallythe accurate information of the target, and make the corresponding operating decision.Containing abundant perceptive information, the AUV vision system has incomparableadvantages beyond other underwater detection system in the independent operation as it isclose to the target. Research on object detection and localization based on the technology ofmachine vision is of significant theoretical importance, and the application value lies in theimprovement of AUV’s independent operation ability and intelligence level.Given the fact that natural light can’t reach or the light is weak in some underwater areas,image there has low contrast ratio for lack of illumination. So image needs underwaterlighting device for better quality. This paper deals with the study of image segmentation,feature extraction, location and tracking of the target using monocular vision for underwaterillumination image. A complete underwater machine vision system for object detection andlocation has been established, to ensure the measurement accuracy, improve the real-timeperformance and stability, so that better independent operation of AUV can be guaranteed.The contents of this research are summarized as follows:Research on object detection method based on monocular machine vision system.Because of low contrast ratio and color distortion of the image with underwater illuminationshows inter-regional connection by using the traditional adaptive threshold method for imagesegmentation. According to this, an image segmentation method based on grayscaletransformation for underwater illumination image has been proposed, this method enlarges thegray level difference between the object and background by changing the weights of threechannels, and reaches the segmentation results by using adaptive threshold method. Thismethod was verified by image segmentation experiment in single sphere object with differentcolors. According to the problems that the object region extraction is incomplete by using thetraditional adaptive threshold method in uneven illumination conditions. This paper proposedan multi-threshold image segmentation method based on sliding window, this method has setup the sliding window in the target area and do the threshold processing to the different area that was covered. The final segmentation results can be achieved by using the differentthreshold segmentation to different regions. According to the problem that sphere objectimage edge has partial deformation in segmentation result, this paper improved the method ofthe traditional Hough transform circle detection to raise the detection precision. Theunderwater illumination image segmentation experiments verified these methods.Research on object three dimensional localization method based on monocular machinevision system. During the research of the object location based on known information,according to the problem that the distortion error in the camera imaging process leads to alower object location precision, this paper add secord-order radial distortion correction in theideal pinhole imaging model, and establish the whole nonlinear camera. During the researchof the object location based on unknown information, according to the problem that singleultrasonic sensor has a narrow neasuring range, a method that obtained the target sizeinformation based on vision guiding is proposed, this method fused the information fromimage sequence and ultrasonic data to obtain the sphere object’s dimension, and then doingthree dimensional object location by using the method of the object location based on knowninformation. The validity of these object location methods is verified by the underwatercontrast experiments.The research on object detection and object tracking based on monocular machine visionsystem. Acooding to the problem that the detection result has a lower accuracy by using thetraditional frame subtraction algorithm to detect object, this paper improved the traditionalframe subtraction algorithm by using spatial characteristic to detect moving target andtemporal characteristic to verify the result. During the AUV go to the object from far and near,gray level distribution in the underwater illumination image is changed because of the changeof light intensity, the inaccuracy result of target tracking is occurred by using the traditionalMean Shift method. According to above, this paper improved the traditional Mean Shiftmethod by adding the spatial information about object gray level and adjust the trackingframe in real-time to improve the accuracy of object tracking during the change of lightintensity.The validity of these object detection and object tracking is verified by theunderwater contrast experiments.