Research On Three-dimensional Point Cloud Model Reconstruction Based On Kinect Sensor

With further research of the computer vision theory and development of softwareand hardware technology,3D object and human body point cloud model has widely usedin scientific research and production. And it is attracting increasingly attention andconcentration. For the expensive price of the traditional3D model building equipment,complex operation and other disadvantages, this paper puts forward the construct3Dobject point cloud based on Kinect method. The contents of this research are as follows:Firstly, elaborated the research background and significance of this article,summarized the three dimensional modeling technology research status at our countryand abroad, introduced the research methods of this article. Discussed the hardware andsoftware of Kinect sensor and its application technology. The depth informationacquisition principle of Kinect and deduce how the depth image transformation to thespatial three-dimensional coordinates, and then elaborated the3D point cloud.Then, analyzed the noise source of Kinect sensor’s image of depth data, and triedmany kinds of filtering algorithm to depth image experiment comparison and analysis.Introduced two different point cloud data object segmentation method, the depth ofbackground difference method and the depth data threshold method..Finally, for the question of the traditional SIFT algorithm operations slower whenget feature points and put forward SURF algorithm to extract feature points of depthimage. Using ICP algorithm to reconstruct3D point cloud image, and build a complete3D point cloud model of the object after wrong match points were deleted byRANSAC algorithm.This paper proposed a new three dimensional point cloud model building method, itcan meet the requirements of reconstruction of the object and human under complexbackground. With Kinect to reconstruct3D point cloud model on common objects,greatly reduce the cost and difficulty of the3D modeling. Eventually the experimentreconstruct3D object point cloud model, it can basically meet the demand of dailyapplication, has a broad prospect of application in practical production.