Skip to Main Content
This paper presents two novel methods to calibrate the 'attitude' of a camera from a long sequence of images captured by a camera mounted on a train. The 'attitude' refers to the camera's orientation with respect to the motion plane. For general plane motion with small rotation such as motion of a train, it is difficult to estimate the 'attitude' of a camera accurately. A long image sequence is used to overcome the noises in rotation estimation induced from the errors in correspondence estimation. We have developed an automated technique to process a large number of images and calibrate the 'attitude' of a camera mounted on the train automatically. Experiments with real indoor and outdoor images have been conducted and the results demonstrated that the methods can estimate the 'attitude' of a camera with good accuracy.