Active Stereo Matching Benchmark for 3D Reconstruction using Multi-view Depths

Abstract

With the advance of 3D entertainment, 3D reconstruction has been widely researched. Recently, for the 3D reconstruction, multi-view depth images are generally used due to the wide availability of commercial RGBD sensors. The depth image can be directly acquired from the specific sensor or estimated from the stereo images by using a stereo matching algorithm. The performance of the depth estimation using a specific sensor is only dependent on the sensor performance. However, since the stereo matching method is dependent on stereo matching accuracy, a more accurate depth can be obtained from the high accuracy stereo matching method. Therefore, we focus on the stereo matching method for estimating the depth image. In this paper, we present the benchmark on the active stereo matching method for 3D reconstruction. Through the quantitative and qualitative benchmarks, we analyze and visualize the depth estimation and 3D reconstruction results. By presenting the active stereo matching benchmark, we provide guidance for 3D reconstruction using multi-view depths.