Accurate Point Measurement in 3DGS -- A New Alternative to Traditional Stereoscopic-View Based Measurements

A team of researchers, Deyan Deng and Rongjun Qin, has published a novel method that unlocks the geometric measurement potential of 3D Gaussian Splatting (3DGS). While 3DGS is renowned for its real-time, photorealistic novel view synthesis, its use for precise 3D measurement has been limited. The new technique bypasses the need for costly stereoscopic hardware and specialized operator training. Instead, users can interact with a standard 3DGS model in a web application, intuitively selecting the same physical point across multiple rendered views. The system then triangulates these 2D selections to calculate an accurate 3D coordinate, mimicking—and improving upon—traditional photogrammetric workflows.

Quantitative tests on UAV aerial datasets demonstrate the method's significant advantages. It achieved root mean square errors (RMSE) in the 1-2 centimeter range for well-defined points. More importantly, it excelled where conventional mesh-based measurements struggle: on a challenging thin structure, the new method achieved an RMSE of 0.037 meters, compared to 0.062 meters for the mesh method. On sharp corners poorly captured in a mesh, the 3DGS-based approach successfully measured all points with a 0.013 m RMSE, while the mesh method failed entirely. This research, accepted for the 2026 ISPRS Congress, provides a compelling, accessible alternative for fields like surveying, construction, and cultural heritage documentation that rely on precise 3D data.