基于RGB-D相机数据的人体三维重建系统

3D Reconstruction System of Human Body Based on RGB-D Camera Data

关于利用RGB-D相机对人体进行三维重建的问题,设计人体三维重建系统,利用连续拍摄多帧的RGB-D图像进行高质量人体三维重建。系统前端提取RGB图像特征点并结合特征点的深度值,通过最小化重投影误差求解相机的运动轨迹,结合深度图像和RGB图像通过融合数据获取初始模型;系统后端构建光照模型,利用RGB图像和模型的亮度变化连续性对人体几何结构提供约束,优化初始模型。算法与主流实时三维重建算法ElasticFusion、BundleFusion作对比分析,表明该文重建算法在人体模型的纹理几何结构恢复方面有比较好的提升。

This work proposes an algorithm for human body 3D reconstruction by fusing multi-frame RGB-D images of the same human body with a hand-held RGB-D camera.The front-end,based on the features of RGB images and the corresponding depth values,allows for accurate trajectory estimation by minimizing the re-projection error and allows for RGB-D frames to get the initial model.The back-end,based on the lighting model,provides the constraints about the brightness continuity of the RGB image and the model on the human model geometry to optimize the initial model.This algorithm is compared with ElasticFusion and BundleFusion.