便携式RGB-D传感器深度测量与建模精度研究

Study on Depth Measurement and 3D Reconstruction Accuracy Based on Portable RGB-D Sensor

摘要:

以RGB-D(彩色-深度)传感器的代表微软Kinect 2.0相机为例，结合飞行时间测距法(ToF)获取深度信息的原理，对环境中不同材质、不同颜色情况下获取的深度数据的精度进行理论与实验分析，得出深度测量误差与深度测量距离的平方成线性关系，以及不同颜色材质下深度测量精度的定量规律.同时利用所建场景、模型表面特征参数指标的绝对误差和相对误差，并对豪斯多夫距离法进行改进，来分析整体建模精度；以高精度三维激光扫描仪(LiDAR)建模成果作为标准，对RGB-D传感器整体建模结果的精度进行评价，在此基础上分析便携式RGB-D在量测、建模上能达到的实际精度，得出其在实际近景测量、室内建模等方面的适用性.

Abstract:

In this paper, we use Microsoft’s Kinect camera as an RGB-D sensor example, analyze the depth error source based on the time of flight(ToF) principle and the obtained depth data accuracy of different colors and materials. We conclude that the depth measurement error is linear with the square of the distance, as well as the quantitative rule of depth measurement accuracy under different color materials. Besides, we use high-precision light detection and ranging(LiDAR) data as reference to evaluate Kinect modeling accuracy through comparing absolute error and relative error of parameters between these two models, and improve the Hausdorff distance method to analyze the overall modeling accuracy. Finally, we show both qualitative and quantitative results relating to the actual accuracy of Kinect depth measurement and modeling and its application scope in close range photogrammetry and indoor modeling.