A Quantitative 3D Reconstruction Evaluation Method Based on Blender

Recently, image-based 3D reconstruction methods have made remarkable breakthroughs. However, the evaluation of 3D reconstruction is still limited to subjective judgment due to the unavailable 3D model data for ground truth. This paper proposes a quantitative evaluation method for 3D reconstruction based on the existing texture CAD models and rendering tools. To this end, we present a data simulator based on the Blender tool to generate multi-view images similar to that captured by a handheld device, i.e. a time-space continuous video of a target object. The data generated by Blender can not only be used to verify the 3D reconstruction algorithms, but also can ensure that the reconstructed geometry has the same coordinate system and scale as the CAD models through the limited camera parameters. Then, we can use the original CAD model to evaluate the reconstructed geometry and color. The experimental results show that our method can effectively evaluate the quality of 3D reconstruction.