Methods And Accuracy Analysis In The Measurement Of Pose Based On Computer Vision

With the development of the theory of computer vision and image acquisitiondevices, measurement method based on computer vision has been widely used inmore and more areas for advantage of non-contact, wide measurement range, lowcost and high automation. The main task of this paper is to design the visionmeasurement system to achieve three-dimensional pose measurement of three-axisair Air-bearing test-bed, the study include overall project designing, cameracalibration, the extraction of corner points of calibration model, design the targetmodel for measurement, position the center’s coordinate of feature points, forecastthe area of feature points, error analyzing and compensation of the extraction offeature points.First, the pose measurement systems of mono-vision and stereo vision areanalyzed from the two aspects of the measurement principle and pose algorithm. Bycomparation of the accuracy and the robustness of pose measurement of the twosystems with the simulation method, the former one is chosen.Secondly, the calibration method by Zhang Zhengyou and ring-pointcalibration method are analyzed and compared, the relationship between the cameracalibration accuracy and calibration distance is studied, and a simple, stable, andbetter robustness X-corner detection algorithm is given out based on the uniquenature of the image gray distribution near the X-corner of the checkerboard.Then, the layout of the feature points is designed from three aspects ofreducing the positioning error of feature points, the unique of the pose measurement,distinguishing the topological relations among feature points easily. The method ofthe center position of feature points is studied an algorithm which the size ofsearch-area is self-adjusting is designed to predict the area of the feature points andimprove image processing speed.Afterwards, random noise sources of the image based on image sensorperformance indicators and the sources of systematic error of the center position offeature point are detailed analyzed. Compensation method of position errors ofelliptical feature points caused by projection is demonstrated.At last, the experimental measurement system of monocular visual is structured; camera calibration experiments and pose measurement experiments arecompleted, and then experimental results are analyzed.