Research On Surface Ship Feature Extraction And Measurement Based On Binocular Vision

Unmanned Surface Vehicle（USV）has great application value and commercial prospects in military,civilian and scientific research fields.It receives much attention from different countries around the world in recent years.Environmental awareness is one of the core contents of USV development,which is particularly important for the vehicles’ safety and stability.This paper combines feature extraction and target tracking algorithms to build a binocular vision measurement system on USV,and provides an effective solution for USV’s environmental awareness.This article briefly introduced the principles of vision measurement technology,and describes the camera imaging model and the conversion relationship between the four coordinate systems in the camera imaging model;then described the camera internal and external parameters of the binocular vision measurement model,and introduced Zhang Zhengyou’s binocular camera calibration Finally,based on Zhang Zhengyou’s binocular camera calibration method,the internal and external parameters of the cameras on left and right were both solved.In order to improve the accuracy of target feature extraction,this paper analyzes the interference of background pixels and image noise on target feature extraction,and uses image histogram equalization to complete the pretreatment of ship images on the water surface to enhance image contrast and reduce image noise interference.Then,the edge of the ship image on the water surface is detected by the Soble operator,and the sub-pixel corner detection is completed based on the Harris detector.Finally,the stereo matching of the binocular camera was completed based on the SURF feature matching operator and the stereo correction result.In order to improve the tracking efficiency of USV on surrounding vehicles,the mean-shift target tracking algorithm was improved.Firstly,this paper analyzed the shortcomings of common target tracking algorithms,and focused on solving the problems of target losing and algorithm cumulative error.Secondly,a SOFM model was established to classify pixels in a rectangular frame and remove background pixels.Thirdly,this paper optimized the EFK solution process,completed the prediction and correction of the target center position,and introduced a similarity function to determine whether the target is occluded.In order to improve the measurement accuracy of the target,the ideal binocular stereo vision model and the actual binocular stereo vision model were introduced.Combined with the operation result of the image processing algorithm,the three-dimensional space coordinates corresponding to the feature points were obtained through equation,and found the distance information.Finally,based on the EKF target positioning algorithm,the real-time speed information of the target was obtained.The research on binocular vision’s methods for the extraction and measurement of vehicle features on the water surface has improved the USV’s ability to sense the environment and ensure the safety of USV’s autonomous navigation.