Research On RGBD Point Cloud Registration Method Based On Illumination Compensation

With the continuous improvement of 3D scanning technology and point cloud model processing technology,computer vision,human-computer interaction,robot navigation control,virtual reality,3D reconstruction and other fields are also making new breakthroughs.However,when measuring an object by a three-dimensional scanner,it is susceptible to changes in the illumination environment,errors in the scanning measurement instrument,or the object itself and other occlusions.The acquired point cloud data has the disadvantages of large noise,large color data deviation,and limited point cloud data acquired from one perspective.Aiming at the limitations of point cloud data acquisition,this dissertation proposes a point cloud registration method for de-lighting interference and a point cloud registration method with adaptive color geometry blending features.Experiments show that the three-dimensional point cloud registration method in this dissertation has improved registration efficiency and accuracy.(1)Aiming at the problem that the illumination intensity unevenness affects the three-dimensional point cloud registration effect,this dissertation studies a RGBD point cloud registration algorithm based on illumination compensation.It is a common image processing method borrowed from two-dimensional images.First,the homomorphic filtering algorithm is used to eliminate the influence of lighting factors on the color data taken by the Kinect device.Then,the color features and geometric features are combined into a hybrid feature,and the K-nearest neighbor(KNN)algorithm is used to search for the corresponding points to improve the search efficiency.The normalized color and geometric coordinate 6D distance metrics are constructed,and the distance between points is selected.Small points are used as corresponding points.Finally,iteratively solves and obtains a rigid transformation matrix.Experiments are carried out by collecting multiple sets of model data under different illuminations.The experimental results show that the proposed algorithm has high stability and significant registration effect in model registration under illumination factor interference.(2)Explore the point cloud registration method using adaptive color and geometric blending features of histogram statistical features.In the first step,according to the number of color types in the source point cloud and the target point cloud data set,the color tolerance is set to remove a few fewer color points,and the points with similar colors are combined and counted.In the second step,two pieces of point cloud to be matched have the same color data as the candidate point set,and then the color and geometric features of the candidate point set are calculated.In the third step,the color and geometric features are proportionally mixed by the histogram statistical method.Then search for the corresponding point pairs,iteratively solve,and obtain the rigid transformation matrix.Experiments were carried out by collecting data of two models with different color and geometric distribution(the color could not be distributed in the geometric concave and convex parts and the color distribution in the geometric concave and convex parts).The experiment proved that the research method was combined with the model of color and geometric data.The registration method has certain obvious advantages.