Fine active calibration of camera position/orientation through pattern recognition

Precise information on an object, which is obtained through pattern recognition, is fused into an active calibration process. The calibration process can be classified into two modes, i.e., passive and active calibration. Passive calibration is possible by using images of calibration grids pre-located at a reference position with fixed size, while active calibration is performed using images of real objects, which are to be manipulated. In most active calibration processes, either based upon a line-correspondence scheme or upon a point-correspondence scheme, clear images of the object are required. Even with uncontaminated images, there exist errors in estimating the position and orientation of a camera due to lens distortion and limited number of pixels, as well as inherent errors due to the perspective model of a camera. These problems are being tackled in various ways since there is no simple and solid approach. In this paper, precise position and shape information of an object, which is obtained by a pattern recognition procedure, is integrated into the calibration process based upon a point-correspondence scheme to improve the accuracy in estimating the position/orientation of a camera. Experiments are performed to show the improvement in this accuracy.