Research On Calibration Technology For Stereo Vision With Large FOV Based On Defocused Image

Stereo vision systems have been widely used in industry inspection and national defense for their features of non-contact,automation,etc.Calibration is the key technology of stereo vision systems,the calibration precision of stereo vision has a critical impact on the precision of measurement.Therefore,researches on high-precision calibration method for stereo vision with large Field of View(FOV)is of great practical significance.In this thesis,a calibration technology based on defocused image has been studied to simplify the operational process and calibrate the stereo visions accurately and steadily.The main works and contributions are as follows:(1)A helicopter blade motion parameters measure system based on the four-view stereo vision has been improved.The hardware components of the system include four high-speed industrial cameras,frequency multipliers,computers,remote control computers and eight photographic lights.The software components of the system includes equipment management module,image acquisition module,stereo vision calibration module,global calibration module,targets three-dimensional coordinates calculation module and blade motion parameters measurement module.This thesis focused on the interface operation and algorithm optimization of the stereo vision calibration module in the original system,improved the operability and stability of the calibration process.After test,the improved system can meet the needs of helicopter blade motion parameters measurement.(2)A method based on defocused image for stereo vision calibration with large FOV has been proposed.First,the defocused images are captured by a small size 2-D template at the defocusing position.The calibration of monocular camera is completed by using the defocused images.Second,the calibration images are captured by a small size 2-D template at the measurement position.The calibration of stereo vision is completed by using the calibration images.In order to capture images efficiently,the field of view area is divided.Finally,the simulation experiments and real experiments are conducted to verify the proposed method.The real experiments conditions are set up as follows: the measure distance is approximately 5.1m,the field of stereo vision is 4.8m? 2.4m?.The experimental results show that the normalized re-projection error and re-construction error are less than 0.23 mm respectively.The RMS measurement error is 0.858 mm,and the relative measurement error is less than 0.3%.Therefore,this method is simple operation,high accuracy and reliable,which can meet the requirements of onsite calibration and achieve high-precision for helicopter blade motion parameters measurement.(3)The factors affecting the calibration accuracy of large FOV cameras has been analyzed.First,the influence mechanism of the defocusing position,the material of template,the size of feature points,the distance between adjacent feature points,the number of defocused images and the noise level on the calibration accuracy are analyzed.Second,under the experimental environment of 5m measurement distance and 4.6m?2.3m field of view,the verification experiment are carried out to test the influence of factors on the calibration accuracy.The experimental results show that: 1)When the defocusing position under the range of 1~2.2m,high calibration accuracy can be achieved,and the best position is 1.4m.2)When selecting the carbon fiber template with a feature point radius of 18 mm,a feature point spacing of 48 mm,and a number of feature point of 9*10,high calibration accuracy can be achieved.3)The method has strong anti-noise ability.4)When the number of defocused images reaches 27,high calibration accuracy can be achieved,but the stability is not good.When the number of defocused images reaches 40,the calibration accuracy and stability are the best.