一种使用角点特征优化KinectFusion的3D建模方法

摘要

3D modeling technology has been widely used in medical imaging,robot navigation,urban planning, virtual reality,terrain exploration and other fields,but how to use the limited processing resources to construct a dense model remains a challenge.This paper presents a 3D modeling method which uses the corner feature to op-timize KinectFusion algorithm.Firstly,each color image captured by Kinect performs corner feature extraction, then the spiral search strategy is used to find the region of interest(ROI)which contains enough corner features, and ICP(iterative closest point)is used to align the ROI with continuous data frames,simultaneous analysis of modeling results for traversable regions.Finally,the experimental results show that the proposed algorithm is al-most equal to the familiar KinectFusion algorithm in modeling accuracy, but its computational speed is about twice as faster as the later.Therefore,this method improves the user experience of depth camera in the field of indoor scene reconstruction,it can complete about two times larger scene reconstruction than KinectFusion meth-od in the same hardware condition.%三维建模技术已被广泛应用在医学成像、机器人导航、城市规化、虚拟现实、地形堪探等领域,而如何使用有限处理资源构建稠密模型仍然是一个具有挑战性的课题.文中提出了一种使用角点特征优化KinectFusion的3D建模方法,首先对Kinect捕获的每张彩色图像进行角点特征提取;然后使用螺旋搜索策略查找包含足够角点特征的感兴趣区域(region of in-terest,ROI);接着对连续数据帧的ROI使用迭代最近点(iterative closest point, ICP);最后,通过实验验证了文中改进方法与KinectFusion算法建模精度相近,但计算速度为其2倍左右.因此,文中方法在一定程度上改善了深度摄像头在室内场景重建领域中的用户体验,在相同硬件设备条件下,可完成KinectFusion方法2倍左右大尺寸场景的重建工作.