Proceedings Article | 21 August 2013
KEYWORDS: Cameras, Visualization, Calibration, Distortion, 3D modeling, Control systems, Mathematical modeling, Optimization (mathematics), Image processing, Precision measurement
Large-size visual shape measurement based on ICP (iterative closest point) mosaicing algorithm generally has a larger
cumulative error; however, this problem can be well solved by precision positioning global control network. Therefore,
this method is widely used in large-size visual shape measurement. Since the positioning accuracy of the global control
network is the key influencing factor of the final measurement accuracy, the method of precision positioning global
control network is researched, which is dependent on the principle of portable close-range photogrammetry. The
precision positioning theory and mathematical model of global control network are investigated in this paper. Bundle
adjustment optimization algorithm is the core of this measurement system, the solution method of this algorithm is
introduced in detail, which can improve the model solution accuracy. As is known, the initial value of the algorithm has a
direct influence on the convergence of the result, so obtaining the initial value is a key part of the measurement system,
including control points matching technology, stations orientation technology and the technology of obtaining the initial
value of the three-dimensional coordinates of global control points. New technological breakthroughs were made based
on the existing researches to get a more precious and stable initial value. Firstly, a nonlinear adjustment model based
control points matching method is proposed, which significantly improves the correct matching rate when the control
points distribute intensively. Secondly, a new station orientation method without using an external orientation device is
studied, which greatly improves the shooting freedom and expands the range of the spatial distribution of the
measurement stations. Finally, a camera calibration method independent with the imaging model is explored, which
converts image coordinate information into image angle information. Thus, the initial value calculation accuracy of the
three-dimensional coordinates of the control points is not affected by the lens distortion and measurement distance. A
large number of experiments were carried out using a high-resolution digital camera, and the experimental results show
that the measurement accuracy of this method can reach 0.02mm (3m * 3m range), and the root mean square is about
0.015mm. Consequently we conclude that this method can achieve the precise positioning of the global control network
and help to improve the accuracy of large-size visual shape measurement.
