Abstract

To improve the digital level of indoor inspection， the reconstruction efficiency and accuracy of the multi‐view geometry reconstruction algorithm based on COLMAP in indoor environments were studied. Seven typical scenes were selected, and 3D reconstruction experiments were conducted based on inspection videos. Indices such as the number of effective images, overlap ratio, reconstruction efficiency and accuracy were defined to achieve comprehensive evaluation of reconstruction results. The results indicate that after fitting based on the power law function, the reconstruction time is proportional to the 1.15th power of the number of effective images. The reconstruction efficiency first increases and then decreases with the image overlap ratio, and the optimal overlap ratio is approximately 70%. The reconstruction accuracy of consumer-grade cameras is centimeter-level. The error in the same scene is less affected by the reconstruction density, and the average absolute （relative） error under the most unfavorable calibration of different scenes is 2.29 cm （3.49%）. The general rules of 3D reconstruction efficiency and accuracy for indoor scene images are revealed, pointing out the direction for more efficient image acquisition and application in indoor inspection.

Download paper here

Download preprint here

The authors are grateful for the suuport of technology project funding from State Grid Corporation of China (No. 5200-202311481A3-2-ZN)

Financial Sources: