Three-dimensional Measurement Technology Based On Structure Light In Binocular Endoscope System

With the large demands of society,the development of three-dimensional measurement technology is very fast,medical diagnosis' stereo analysis has become important branch of computer vision.After acquiring three-dimensional information in vivo by images through electronic endoscope,doctor can view and measure lesions tissue clearly,which are able to improve success ration in surgical operation,decrease complications after operation and shorten recovery cycle.This thesis is developing a three-dimensional measurement system based on binocular endoscope,which can visualize and measure object surface.The main work is to design software system with binocular calibration and stereo matching algorithm achievement,also develop and test binocular system based structure light.Firstly,according to the demands and system's performance,we design overall scheme and integrated software.The software will be presented in binocular images display,cameras calibration and object's depth map display and the measurement function.Finally,we prove the system's feasibility through experiments.Achieving image pair undistortion and pole alignment by using Zhang's calibration algorithm,in order to get binocular system's internal and external parameters,we develop manual and automatic camera calibration program.Through analyze the state of the art stereo matching algorithm,and balance the match speed and accuracy,we select the semi-global algorithm.In order to deal with weak texture or duplicate texture area,we propose active structure light illumination based on binocular system.In practice,we use laser and ground glass to generate speckle,which can compact the system structure,as well as provide the features on object surface.Under the condition of processor clock speed is 2.5GHz,binocular system base line length is 2.7516mm,and CMOS resolution is u,we use green laser to hit the ground glass,and experiment on different material objects with weak texture.The results show this method can achieve good performance,and the system has relative measurement accuracy of 1.03%in the distance of 12mm,as well as the speed can achieve 18 frames per second.