Coded points can be used to resolve the accumulative error in registration and the matching complexity of the uncoded points in large objects’ 3D measurement is greatly reduced.An self-adaptive binary method is adopted to segment image

and the reference points are extracted based on their geometric and grayscale characteristics.Then the coded points and uncoded points are classified.By choosing the point with gray jumping in coded points as the start point to be decoded

and using both the length and its central angle of the arc

the decoding accuracy of the coded reference points is improved.Both simulation and experiments show that this method is robust to noise

its correct decoding rate is up to 97%.