Algorithm Research Of Third Point Cloud Registration Based On Feature Matching

With the requirement of production efficiency and product quality in modern industry, computer aided manufacturing engineering has become an important component of modern computer science.3D point cloud registration algorithm Is an important branch of computer aidedmanufacturing engineering, by which can rise up control industrial production accuracy of measurement error and speed up the industrial production schedule, etc.3D point cloud registration algorithm is in order to get a complete measurement object, Itis measure angles of point cloud data information for information integration In principle. The core issue of it includes two aspects, First, whether to stay from registration of point cloud data accurately extract the corresponding feature points each other in the set. Second, if we caneffectively reduce the noise impact on point cloud registration information. According to the key problem mentioned above, this article carry on research mainly from the following three points.Three-dimensional point cloud registration algorithm is in order to get completemeasuring objects, and coordinate axis of the measured point cloud data from multipleperspectives. Its core problem includes two aspects: first, whether we can extract the accuratearea characteristics from point cloud data. Second, if you can reduce the effects of noise onpoint cloud registration information. The two problems emerge in the entire research process.This article mainly studies third point cloud registration algorithm from three different levels,and are based on the feature matching algorithm:The first level, studying third reconstruction of2-d image sequence, and is conversionfrom two-dimensional information to third-dimensional information. This paper proposes aimproved algorithm and can improve the accuracy and timeliness of image stitching undercertain conditions. First, we should take into account two stitching image of characters not easy match in case of the visibility and shooting Angle photographs different. The part ofimage smoothing Gaussian filtering is used to eliminate the noise points. Second, In view ofthe feature points extraction is not easy to match or match failed, we can make the samecharacteristic regional markers, and thus subject to SIFT through the relationship betweendistance and Angle to eliminate algorithm to extract the two figure does not match the featurepoints, reduce subsequent matching time and improve the success rate of image stitching.The second level, research what industrial measuring accuracy evaluation oriented sparsepoint cloud registration method and sparse three-dimensional point cloud registrationalgorithm. Industrial manufacturing of precision have higher requirements, the need forprecision measurement to evaluate the product under test is in line with the industry standard.This paper proposes the improved sparse point cloud registration method. First, match twopoint cloud by principal component analysis (PCA), and make the space coordinates them one.Then, determine whether two recent cloud point s iterative mean square error is less than thesetting threshold. If not, adopted at high contact ratio matching point by random sampling theconsistency (RANSAC) algorithm, and make the next point cloud registration, or match end.the adoption of RANSAC algorithm to extract the high contact ratio on match point, ensure toget the optimal space coordinate transformation parameters, makes the point cloudregistration precision is higher, and the estimate of the number of sampling can get areasonable point cloud registration number of iterations, achieve the goal of reducingoperation time.The third level, researching dense third point cloud data registration. This paperconsiders two aspects: the first is the Kalman filtering process that has a good noise reductioneffect to dense third point cloud data and the noise reduction of point cloud registration willbe more accurate. The second is many new interest theory can predict and update forregistration of point cloud data by retain enough valid information. Combining these twoalgorithms, this paper forecasts and updates point cloud data by many new coupon Kalmanfilter, and get more efficient point cloud data. Providing more accurate transformationparameters for the point cloud. General algorithm process is: first use PCA coarse registration algorithm makes the basic coordinates corresponding to match on time cloud data, and thenthe coarse registration after the point cloud data to many new rates kalman filteringprocessing, finally using the essence of the ICP registration algorithm for many times, madewith on time cloud data for better matching effect.