The Research Of Real-time3D Reconstruction In Mobile Scenery Based On Point Feature Histogram And Information Fusion

With the development of SLAM technology and3D imaging equipment, theresearch of three-dimensional reconstruction in mobile scenery is getting hoter. Themain research of this paper is that mobile robot carrying sensors move in a scene,through continuous scanning, It can reconstruct the whole scene in real-time. In order tosolve this problem, the paper’s main work is as follows:（1）In mobile secnery, the quantity of the points is large and there exists complexnoise,algorithms need to be real-time and must keep enough geometrical features, thetraditional simplification methods based on distance, curvature and feature have beenunable to meet the demand of new applications. This paper introduces a new featuredescriptor-Point Feature Histogram. As the points’ PFH distribution exist significantdifference on the different geometry surfaces, we presents a new simplification methodin mobile scenary based on PFH.The algorithm uses the standard deviation of PFHdistribution to classify the points quickly, after compression,enough geometricalfeatures of the point cloud can be reserved and the compression rate can also meetrequirement.（2）Multi-source image fusion is a method that use certain algorithms to combinedifferent images obtained by two or more senors at the same time or different time toget better discretion information about the scene. This information can’t be obtainedfrom a single image source.Through the fused image we can obtain more reliable,richer, and more useful information, these information is more suitable for the computerto detect, classify, identify, understand, and process. We come up with a fusionmethod based on regional mask.The use of the fused image can improve the speed oftarget identification and target tracking, and significantly improve the accuracy ofrecognition. The fused image contains not only depth information, but also contains richsurface texture information. We can reconstruct more realistic three-dimensional sceneby using the fused images.(3) In this paper, we proposed a real-time3D reconstruction method in mobilescenery based on PFH and information fusion. On one hand,we use special algorithm toreduce the number of the input point set of the ICP algorithm. By adopting the idea ofinformation fusion, we use the offset value of inertial sensors to provide an effectiveinitial value for ICP algorithm,thus improving the convergence speed and the real-timeperformance of the whole reconstruction algorithm. On the other hand, in order toimprove the accuracy of the reconstruction algorithm, this paper also uses the thought ofimage fusion, through the RGB image to repair the surface texture of the reconstructionmodel, thus obtaining more accurate reconstruction results.