Research On Front-end Image Processing Method Of Underwater Visual SLAM

Unmanned Underwater Vehicle(UUV)is a kind of navigation equipment that can carry out underwater tasks independently of human intervention.Simultaneous Localization and Mapping(SLAM)is a mobile robot technology.Through the analysis of sensor data,this technology can simultaneously obtain its own position and build the environment in an unknown environment.However,the images have their own defects such as high noise and poor texture,which makes it challenging for UUV to realize SLAM algorithm using camera.In order to improve the stability and efficiency of underwater SLAM method,this thesis studies the methods of image quality enhancement and image feature extraction based on underwater images.The main research contents as follows:Firstly,an underwater image enhancement algorithm based on multi-input fusion is proposed to solve the problems of underwater image degradation and serious interference.By analyzing and comparing the performance of white balance,histogram equalization,gamma correction and underwater dark channel prior methods,an image enhancement algorithm is obtained by combining the advantages of each algorithm.The algorithm is compared with the conventional underwater image enhancement algorithm,and the image is analyzed by image effect and image quality evaluation index.The results show that the proposed algorithm can effectively improve the underwater image quality.Secondly,aiming at the problem of less evaluation of underwater image feature detection methods,the conventional underwater image enhancement algorithm and the enhancement algorithm proposed are used to process underwater images,and the obtained images are compared with the original images.The performance of six feature detection methods,namely Scale Invariant Feature Transform(SIFT),Speeded Up Robust Features(SURF),Oriented fast and Rotated BRIEF(ORB),Binary Robust Invariant Scalable Keypoints(BRISK),AKAZE and SuperPoint.is analyzed and evaluated.The analysis results show that ORB has the best feature effect.Thirdly,the light source used by the underwater camera will lead to uneven illumination in the underwater image,resulting in low pixels in most images.ORB features are also affected by image pixels.Based on the problem of low pixels,a dynamic mask method is proposed.The mask threshold is set according to the frequency of image pixel value,and the image mask is processed to improve the efficiency of image feature extraction.The analysis results show that the dynamic mask method makes the image have a good effect on ORB feature extraction.Finally,it uses ORB-SLAM3 algorithm framework to carry out simulation experiments.The original image,the image enhanced by the algorithm and the image processed by the dynamic mask method are input into the ORB-SLAM3 framework as sensor data,and the positioning and composition effects are compared with the angle of image feature matching and the angle of simulation trajectory.The analysis results show that the combination of image enhancement and dynamic mask method has the best simulation effect.