Research On Vision-based AUV Recovery End Positioning Method

The Autonomous Underwater Vehicle(AUV)is an important tool for exploring the ocean today because of its high autonomy,high concealment and high flexibility.Limited by underwater communication and its own battery life,the AUV needs to be docked with the recycling dock to complete data transmission,ordering,and energy replenishment.Therefore,as an autonomous robot,how to autonomously locate the recycling dock and approach it is an indispensable part of the research.Due to the limited range of GPS signals under water,acoustic positioning and electromagnetic positioning are limited by the environment and their own accuracy,and cannot meet the needs of recycling end positioning.Trying to apply machine vision to underwater positioning to assist AUV recycling has become a hot research topic in recent years.According to the work flow of underwater visual positioning,this paper studies the algorithm of each step,and puts forward a set of underwater visual positioning strategy for light source target.Starting from image preprocessing,this paper first analyzes the decay characteristics of light propagation in water,on the basis of this characteristic,an image restoration method based on the degradation model in water is proposed,and according to the characteristics of the target image with light source,the image Enhancement and filtering are used to highlight the target in the picture.On this basis,the OTSU algorithm is improved to enable it to segment the light source target and suppress invalid information more effectively.On the basis of extracting the target area,in order to distinguish whether the extraction target area is the target light source,this paper uses Alex Net as the feature extractor and SVM as the classifier,and proposes a light source recognition algorithm based on CNN features.For binocular visual positioning,the extracted target needs to be matched first.This paper analyzes the principle of binocular stereo matching and combines the extracted light source features to improve the constraints of binocular stereo matching to improve the accuracy of matching and real-time.Then,in order to extract the center of the target as the positioning target,this paper improves the Hough transform and adds multiple constraints according to the characteristics of the light source target and the previous processing results,making it possible to use the Hough transform to locate the center of the light source target;for research the principle ofvisual positioning,this paper then analyzes the imaging model of the camera,explains the principle of using Zhang Zhengyou's calibration method,and analyzes and derives the underwater calibration imaging model with refraction parameters to provide a theoretical basis for the improvement of the following algorithm;After obtaining all the positioning prerequisites through the above method,this article immediately analyzes the principle of the traditional binocular vision model,and improves the positioning model according to the actual underwater shooting conditions,and derives the underwater binocular vision positioning model to compensate for the refraction As a result of positioning errors.Then,based on the array of docking light source targets,a monocular visual positioning model was designed as a complement to binocular visual positioning.Finally,this paper designs and conducts the camera on-land and underwater calibration experiments to find the internal and external parameters of the camera,which provides the prerequisites for single and binocular positioning.Then,based on the calibration data,we designed and conducted a vision-based AUV recovery end simulation target positioning experiment,analyzed the positioning data and characteristics obtained by different positioning algorithms in this paper,and verified the feasibility of this algorithm and visual positioning strategy.