Key Techniques Of Feature Points Photographed Measurement

Based on the demand and application of3D feature point measurement, thispaper analyzed and compared the principle and character of the system of typical3D feature point measurement at home and abroad, and researched some of keytechniquese in TNFP system.This paper analyzed the character of self-calibration bundle adjustmentapplied to aerial photography measurement and3D feature point measurement.As the nets of measurement are flexible in3D feature point measurement, thebanded structure is not included in coefficient matrix of nomal equation thatformed with bundle adjustment. Thus, computing speed based on generalizedinverse method is slow. To avoiding the problem, a new fast algorithm ofself-calibration bundle adjustment based on the simplification of the normalequation which using collinearity equation of R-T-mode is researched. Firstly, tobuild the coefficient matrix of normal equation of camera parameters, followedby calculating its corrections. Secondly, to calculate the3D corrections bypoint-to-point, thus avoiding the formation of global error equation and normalequation of3D point. Some experiments validated that computing speed can beimproved efficiently without loss precision, and particularly suitable for TNFPsystem with angle varied shoots.According to the character of TNFP system, a camera self-calibrationtechnique based on factorization and bundle adjustment was described. With thehand-held and fix-focus camera undergoing at least three arbitrary motionsaround the calibration pattern, all the intrinsic parameters and the distortioncoefficients can be obtained. The proposed method has three novelties. Firstly, itsrobustness is markedly increased since all the images are aligned in thefactorization process. Secondly, the non-linear optimization algorithm bundleadjustment guarantees high accuracy. Thirdly, the proposed method does notrequire specialized calibration pattern or rigid camera motion, which make it beused in a wide range of applications. Some experiments validated that4-parameters optical distortion model can be used to compensate the systematicerrors of digital camera, and particularly suitable for TNFP system based on image sequences.On the basis of decoding theory of coded point, a new removed methodabout error decoding of mark point was proposed, thus, it is better to eliminatethe influence of the mismatch of mark point by effective. TNFP system has beensuccessfully applied to industrial inspection, and when the range of measurementis within4000mm×2000mm×2000mm, the measurement precision could besuperior to0.05mm/m.