K.K. Chiang and K.L. Chan (PRC)
Calibration, scale factor, linear and non-linear optimisation
This project aims to implement 3D object reconstruction from an image sequence acquired from a mobile digital camera. Each camera view contains both the object and the calibration pattern. Camera calibration can be carried out independently on each of the acquired views under different camera settings. This provides maximal flexibility for image acquisition. Experiments are carried out to estimate the camera parameters using linear and non-linear optimisations. It is well known that linear calibration is computationally efficient but the solution is sub-optimal. Non-linear calibration yields optimal solution but it is computationally complex. This project is stimulated by the facts that the linear calibration methods do not estimate some intrinsic parameters whilst the success of the non-linear calibration methods is influenced by the starting estimates of camera parameters. However, there have been very few techniques available to solve the problem and all of them cannot be implemented without sacrificing the simplicity and flexibility of the system. Therefore, thorough study has been carried out in the initialisation of those parameters, in particular the scale factor. A novel technique is developed which calculates the scale factor from each acquired view of the calibration pattern. Our results show that the initialisation of the scale factor not only significantly improves the linear calibration method, but also produces good solution in the non-linear calibration method as compared with a fixed scale factor.
Important Links:
Go Back
