Terrestrial SfM-MVS photogrammetry from smartphone sensors

Abstract

Smartphones can be regarded as cameras, natively equipped with geolocation and orientation sensors, making them powerful, portable, user-friendly and inexpensive tools for terrestrial structure from motion/multiview stereo photogrammetry (SfM-MVS) surveys. Camera extrinsic parameters (i.e. camera position and orientation), required to produce fully georeferenced SfM-MVS 3D models are available for the majority of smartphone images via inbuilt magnetometer, accelerometer/gyroscope, and global navigation satellite system (GNSS) sensors. The precision of these internal sensors is not yet sufficient to directly use them as input to SfM-MVS photogrammetric reconstructions. However, when the reconstructed scene is significantly greater than the positional error, camera extrinsic parameters can be successfully used to register 3D models during post-processing. We present a survey of a 400 m wide vertical cliff to illustrate a workflow that enables the use of smartphone cameras to generate and fully georeference photogrammetric models without employing ground control points. Survey images were acquired at a distance of ~350 m to the mapped scene using a consumer-grade smartphone. This survey image dataset was subsequently used to build an unreferenced 3D model, which was registered during post-processing using orientation and position metadata tagged to each photograph.