Document Type : Original Research Paper

Authors

• M. Shafiei
• A. Milan
• A. Vafaeinejad

Department of Surveying Engineering, Faculty of Civil, Water and Environmental Engineering, Shahid Beheshti University, ‎Tehran, Iran

Abstract

Background and Objectives: The integration of imaging technologies with geolocation sensors such as GPS, accelerometers, gyroscopes, and compasses enables precise determination of the camera’s position and orientation during image capture. This capability plays a crucial role in simplifying the georeferencing process of 3D models, particularly in mobile mapping systems and short-range terrestrial applications. In this context, smartphones, equipped with these geolocation sensors, have gained significant prominence as imaging devices. This study examines the feasibility of direct georeferencing of 3D models generated through Structure-from-Motion (SFM) photogrammetry using camera position and orientation data simultaneously recorded by the sensors embedded in a smartphone.
Methods: The research data comprises ground control points collected using a Leica TS09 R1000 total station in reflector less mode, along with data obtained from an iPhone 13 Pro smartphone. In the proposed method, the relative orientation parameters estimated by the Structure-from-Motion algorithm were refined using orientation parameters directly measured by the smartphone’s motion sensors. Image acquisition was conducted under static conditions; therefore, the estimated and measured angles at each station should remain constant and, in the absence of errors, should be identical. The final model was oriented using the average of the measured angles at each station. The measured distance was then used to establish the model scale, and the average coordinates obtained at each station were employed to transform the model into the UTM coordinate system. Additionally, an alternative model was generated using the precise positions of the imaging stations via indirect georeferencing. The RMSE of the check points was used as the accuracy assessment metric for the proposed method.
Findings: The studied stockpile was approximately 150 meters in length and 25 meters in height, oriented roughly from south to north. In this study, five 3D models were generated based on the collected data. The first model was constructed solely using images, positional data, and angles recorded by the smartphone’s sensors. Subsequently, coded targets were incrementally added as control points, with one, two, and three control points being incorporated into the equations. The fifth model was generated using the same images but with the precise coordinates of twelve imaging stations measured by a total station.
In the first model, the positioning error remained at the accuracy level of the iOS single-point positioning system (approximately 2 meters in planimetric accuracy and 11 meters in height). In the model with a single control point, aligning the model with the low-accuracy GPS coordinates from the smartphone resulted in an azimuth calculation error, leading to model rotation. Although the model was correctly transferred, the azimuth error caused a misalignment with the reference system, resulting in an error of approximately 0.40 meters. With the addition of two and three control points, accuracy improved, reaching 0.004 meters, which matched the accuracy achieved using precise camera center coordinates. In the fifth case, where the exact coordinates of the imaging stations were used, an accuracy of 0.0007 meters was obtained.
Conclusion: As observed, GPS accuracy remains the most challenging aspect of this system. Most smartphones do not provide raw GNSS data or high-quality raw positioning data; instead, they only offer the final processed position determined by the operating system. Given the hardware and environmental limitations, the following considerations are recommended for future studies:

a) Development of a handheld mobile mapping system by integrating survey-grade GPS with a smartphone camera.

b) Since the study area consists of earthwork features with irregular shapes, when applying this method to structured objects such as buildings, it is recommended to refine the relative exterior orientation parameters of images using constraints derived from prominent vertical and horizontal features in the scene and incorporating them into the bundle adjustment process.

Keywords

• Smart Phone
• Direct Georeferencing
• Close Range Photogrammetry
• Structure From Motion

Main Subjects

• Photogrammetry