Research On Feature Extraction And Matching In Videometrics

With the development of science and technology， large-scale construction’simportance increases rapidly. As a direct token of mechanics, industry products andengineering structure base on object’s surface and deformation’s measurement with highprecision. The requirement performates all the periods of produce, and during thedesigning period, high-precision measurement is a precondition of converse engineer.The traditional3-D touching measurement techniques has disadvantages ofslow-speed， small-range and easy to maneuver， sometimes even unable to completethe measurement. Videomatric techniques has become a hotspot of research anddevelopment due to its focus detection speed， large measuring range， high precisionand automated processing of data. The camera measurement based on the principle ofbinocular stereo vision computes3-D geometric and movement parameters of thedetected objects through the image snapshoted by the camera imaging system. Ingeneral， the camera measurement can be divided into camera calibration， imageprocessing and solver. By positioning same characteristics on different images， restorethe2-D image information into a3-D position information, in which carema calibrationand solution of parameters deeply rely on image feafures’ extraction. Thus image featureextraction with high-precision is quite important.This paper focuses on the movement measurement and the large-scale structuremeasuring feature extraction and matching algorithms. Point features include naturalfeature points and cooperation marked point, and cooperation marked point includescoding point and non-coding point.This paper studied the structural characteristics of natural extraction methods.Thispaper introduces Harris-algorithms Harris-Laplace algorithm for scale and affinetransformation corner. Harris-Laplace searchs corner in the scale space to ensure thatthe in different scales the image taken by the camera on the same object can be detectedat the same point.. For spots, this paper introduces the SIFT algorithm. SIFT usedGaussian differential operator in scale space to detect spots， and generated128theVader local image descriptor for matching. For corner extraction of Harris, this paperproposed a method that using SIFT to assist Harris algorithm, which is valid byexperiment. The proposed method can hurdle illumination, scale, rotation and noise.This article describes features of circle，vertical angles and crosshairs， threenon-coding cooperation logo compared with traditional methods， and the improvedalgorithm for each sign according to characteristics. For circle features, RandomizedHough transform can be used to detect round flag based on edge information， whichcan save much time； For corner features, Gray characteristic can be used to detect vertical angles through establish gray criterion， which improves speed and accuracy.Crosshairs edge gradient information can be used to edge cluster， reducing the amountof computation to improve the accuracy.This paper particularly introduces rectanglar and circle features’ generation anddetection. Improve precision through adding location circle in circle coding symbol.The two symbols has the advantages of simple structure and high quality imaging,which can be applied in engineer. This paper proposes a method for calculating basicmatric by matching results and retorting matching results by epipolar line.This paper researchs on feature extraction and matching, valided the applicablilyand parameter effection of existing natural construction feature and cooperativesymbols’ extracting and matching algorithm, and propose an ameliorate algorithm toimprove its accuracy and area of application. The algorithm has been tested byexperiment. This article provides project, theory and data basis for engineer. This articlecan provide consults for the extraction and matching research.