Eunju Kwak

Precise Photogrammetric Reconstruction Using Model-Based Image Fitting for 3D Beam Deformation Monitoring

Periodic structural health monitoring of infrastructure systems is important to avoid economic losses and human casualties. Traditionally, deformation monitoring has been done through surveying techniques. Recently, with the increased availability of inexpensive off-the-shelf cameras, photogrammetry has become a viable noncontact alternative for complete three-dimensional reconstruction of the object or surface of interest. This paper aims at combining two methodologies of photogrammetric reconstruction—image-matching-based reconstruction and model-based image fitting—to achieve submillimeter precision for the estimation of both vertical deflections and horizontal displacements. The proposed methodology was tested with data collected using a photogrammetric system at a structures laboratory where a concrete beam was subjected to different loading conditions by a hydraulic actuator. The experimental results showed that the photogrammetric system was capable of monitoring both static and dynamic deformations. The methodology used exhibited a high level of automation and the final results yielded a root-mean-square error (RMSE) of half a millimeter.

Stability Analysis for a Multi-Camera Photogrammetric System

Consumer-grade digital cameras suffer from geometrical instability that may cause problems when used in photogrammetric applications. This paper provides a comprehensive review of this issue of interior orientation parameter variation over time, it explains the common ways used for coping with the issue, and describes the existing methods for performing stability analysis for a single camera. The paper then points out the lack of coverage of stability analysis for multi-camera systems, suggests a modification of the collinearity model to be used for the calibration of an entire photogrammetric system, and proposes three methods for system stability analysis. The proposed methods explore the impact of the changes in interior orientation and relative orientation/mounting parameters on the reconstruction process. Rather than relying on ground truth in real datasets to check the system calibration stability, the proposed methods are simulation-based. Experiment results are shown, where a multi-camera photogrammetric system was calibrated three times, and stability analysis was performed on the system calibration parameters from the three sessions. The proposed simulation-based methods provided results that were compatible with a real-data based approach for evaluating the impact of changes in the system calibration parameters on the three-dimensional reconstruction.

Comparative Analysis of Different Approaches for Multi-camera System Calibration

Mobile mapping systems integrate a set of imaging sensors and a position and orientation system. To fully achieve the accuracy of the utilized sensors, a careful system calibration should be carried out. The system calibration involves individual sensor calibration and the mounting parameters calibration relating the system components. For multi-camera systems, the mounting parameters involve two sets of relative orientation parameters (ROP): the ROP among the cameras and the ROP between the cameras and the navigation sensors. This paper proposes a mathematical model for a single-step photogrammetric system calibration suitable for both single and multi-camera systems. As a special case of this model, indirect geo-referencing can be performed with relative orientation constraints (ROC). A general model which allows the estimation of ROP as wells as the incorporation of prior information on the ROP among the cameras during the integrated sensor orientation (ISO) is also proposed. This general model can be used for the indirect georeferencing with ROC as well as ISO without ROP among the cameras. To evaluate the performance of the single-step system calibration using the different models, real dataset captured by a hand-held multi-camera system is used. The system calibration is performed by using five different models and the performance is compared among these models.

Implication of the number of utilized images on the quality of generated building models using model-based image fitting

The importance of digital building models and their necessities in many applications have attracted the focus of research efforts in the photogrammetric and computer vision communities. Therefore, many research activities have been attempted using different data sources and processing strategies to increase the automation level and accuracy. Limitations of using data from single sensors and the rapid development of sensor technology encourages combining data from multiple sources. Thus, this paper aims at generating accurate digital building models using airborne LiDAR data with imagery by combining two processing strategies - a data-driven approach and a model-based approach. In this research, initial rectangular models are derived from LiDAR data and model parameters are refined through model-based image fitting procedure. The advantage of the proposed approach is that it can utilize any number of images and overcome the limitation of occlusions which are the most common problem when dealing with large scale imagery over urban areas. The proposed methodology is tested using different number of images and the results demonstrate that the accuracy of the models are consistent regardless of the number of images used, even when a single image is used.