Comutations underlying the measuremnt of visual motion.
Artificial Intelligence
Estimation of Object Motion Parameters from Noisy Images
IEEE Transactions on Pattern Analysis and Machine Intelligence
Dynamical Systems and Motion Vision
Dynamical Systems and Motion Vision
Relative Orientation
Digital image processing of earth observation sensor data
IBM Journal of Research and Development
Analysis of linear interpolation schemes for bi-level image applications
IBM Journal of Research and Development
3-D Motion and Structure from 2-D Motion Causally Integrated over Time: Implementation
ECCV '00 Proceedings of the 6th European Conference on Computer Vision-Part II
| Hi-index | 0.00 |
Motion vision deals with the analysis of image sequences acquired by a camera moving relative to an environment. The goal is to recover the motion of the camera as well as the structure of the environment. We model the changing structure of the scene perceived in the camera as a dynamical system. The state of this system is the depth, the distance of a scene point from the camera, the measurement is the optical flow, a vector field of image velocities which can be computed from the image brightness arrays acquired by the camera. We use the dynamical systems model in the construction of a Kalman filter which optimally estimates the depth of a point in the image incrementally over the sequence of frames. By using one Kalman filter per image pixel we are able to recover a dense depth map of the environment, i.e. the scene structure. In every filter iteration we estimate the motion parameters of the camera from the optical flow field using the filter's current depth estimate in a least-squares fashion. The resulting algorithm recovers both the camera motion and the scene structure in an systematic way which could not be accomplished previously. We have implemented this algorithm and tested it on real and synthetic images.