Precision mapping through the stereo vision and geometric transformations in unknown environments

Geodetic techniques for surveying and rapid mapping need to be revisited due to the present progress on satellite, sensor, and geospatial technologies. Conventional surveying methods provide high level of accuracy but require significant human involvement in the field while GNSS (Global Navigation Satellite System) positioning method, provides unsatisfactory accuracy in urban or high vegetated areas due to the degraded GNSS signal coverage. In this study, an alternative surveying method is proposed which facilitates the process of characteristic point localization, using stereoscopic vision and at least one visual marker. At first, the camera system localizes itself and maps the study area using stereo SLAM (Simultaneous Localization and Mapping) algorithm while subsequently detects the visual markers (origin and targets), placed in the area. Afterwards, using a multi-view geometry method for the marker localization and an optimization algorithm for origin marker's plane alignment, the system is able to export the coordinates of the markers and a point cloud (provided by SLAM) in a local coordinate system based on the origin marker's pose and location. The study involves both terrestrial and unmanned aerial vehicle platforms that may carry the proposed equipment. An extensive set of indoor and outdoor, terrestrial and UAV experiments validates the methodology which succeeds a horizontal and vertical error in a level of 10 cm or better. To the best of our knowledge this study proposes the first surveying alternative which requires only a stereo camera and at least one visual marker in order to localize specific and arbitrary points in a centimeter level of accuracy. The proposed methodology, demonstrates that the use of low-cost equipment instead of the costly and complicated surveying equipment, may prove sufficient to produce an accurate 3D map of the scene in an unknown environment.

Automated summary

In this study, a surveying method that combines stereoscopic vision and visual markers is proposed, which achieves centimeter-level accuracy in localizing specific and arbitrary points. The method utilizes stereo SLAM algorithm to localize the camera system and map the study area, and then detects visual markers placed in the area. Using a multi-view geometry method and an optimization algorithm, the coordinates of the markers and a point cloud are exported in a local coordinate system. Extensive experiments validate the feasibility of the methodology, achieving horizontal and vertical errors of 10 cm or better.