基于GPU的三维人脸数据动态线性快速修复

基于CUDA和Open GL,根据三维人脸的深度与灰度数据构造三维点云模型,在更短的时间内显示具有逼真效果的三维人脸模型,并根据传统三维人脸点云显示对数据采集要求较高的缺陷,提出了一种用于修补低质量三维采集数据中出现的空洞现象的动态线性修复方法并使用GPU为算法进行了加速。该方法从人脸数据集中读取点云数据,逐行扫描寻找点云数据中出现的空洞并使用符合人脸本质特征的线性插值方式修补空洞。实验证明,该方法能得到远好于传统点云显示的效果,且经过GPU加速后显示效率得到大幅提高。

Fast linear recovering algorithm for low quality 3D face data based on GPU

In this paper, the modeling efficiency is highly improved by using GPU(CUDA), OpenGL to construct the 3D pointcloud model, and the model is realistic. To resolve the problem of traditional point cloud display which requires high quality database, this paper proposes a dynamic linear recovering method to recover the holes which occur in 3D pointcloud model if it is based on low quality 3D data, and then use GPU to accelerate this process. This pointcloud data is inputed from the 3D face database and scanned progressively for holes occuring in the pointcloud, and the linear interpolating method which is consistent with human facial intrinsic feature to recover the holes is also used. The experiments show that the display effect could be far better than that the traditional point cloud display by our method, and the display efficiency can be greatly improved by using GPU acceleration.