Incorporation of unreliable information into photogrammetric reconstruction for recovery of scale using non-parametric belief propagation

Abstract

The creation of large photogrammetric models often encounter several difficulties in regards to geometric accuracy, scale and geolocation, especially when not using control points. Geometric accuracy can be a problem when encountering repetitive features, scale and geolocation can be challenging in GNSS denied or difficult to reach environments. Despite these challenges scale and location are often highly desirable even if only approximate, especially when the error bounds are known. Using non-parametric belief propagation we propose a method of fusing different sensor types to allow robust creation of scaled models without control points. Using this technique we scale models using only the sensor data sometimes to within 4% of their actual size even in the presence of poor GNSS coverage.