Orientation Parameters Estimation for Uncalibrated Close range Image Sequences Without Scene Information

Abstract:

Metric reconstruction is the highest level reconstruction obtained from uncalibrated image sequences without any scene information, which needs recovering interior and exterior parameters. According to the stratified reconstruction approaches in computer vision, projective matrix and structure are recovered from the correspondences of more than three views through iterative singular value decomposition algorithm, then interior parameters can be estimated through the absolute quadric self calibration method, then exterior parameters are recovered subsequently. With this approach, free models are available without any scene information, Euclidean reconstruction can be realized through absolute orientation when sufficient scene controls are available. The orientation results can be used in low accuracy 3 dimensional measurement or as the initial values for high accuracy bundle adjustment.