Recon-3D Measurement Accuracy Study for Small Scenes
Eugene Liscio & Jihwa Lim
Abstract: Recon-3D is an iOS mobile application dedicated to crash and crime scene documentation of small scenes, which fuses the Light Detection and Ranging (LiDAR) sensor and video frames to reconstruct 3D geometry as point clouds. In a recent training course, sixty students were asked to set up a mock scene with numbered measurement markers, scan the scene with their mobile devices, and provide 10 measurements between the numbered measurement markers in their scenes (n = 600). The results of these measurements were compiled and tabulated for accuracy. The average error of all participants was found to be approximately −2mm with a standard deviation of 15mm. The mean absolute error was found to be approximately 1cm and the maxi- mum error for any one participant was 10cm. Expressing the errors in terms of percent, the average error for all participants was approximately −0.078% with a maximum percent error of 2.83%. Although these measurement exercises were uncontrolled, they show that the majority of errors (2σ), fell within 3 cm. Future studies using point-to-point measurements should include repeatability tests in a controlled environment as there were several variables which were unaccounted for in this study.
PDF: 2023-Recon-3D Measurement Accuracy Study for Small Scenes-Liscio
Supplemental Data: 2023-Supplementary Appendix A-Liscio