## Frisby. 3d scene segmentation using a constrained motion parameter recovery algorithm

- Technical report,
- 1991

2 Excerpts

- Published 1992 in BMVC

A method of motion parameter estimation for AGV's is developed based on a new trajectory constraint algorithm. It is assumed that the vehicle will follow a circular path over the vision sampling interval. The algorithm has been found to be more effective and consistent than least squares estimators when the motion obeys the trajectory constraint. Reliable estimates of the motion parameters can even be made from individual data pairs. In this way, a static world can be segmented from moving objects, and so the motion parameters can be obtained using the stationary points alone. In practice the vehicle will not necessarily follow a circular path, and hence there may be a bias in the parameter estimates. Experiments were carried out using simulated data, where the true trajectory was a clothoid, to investigate the robustness of the algorithm when the trajectory constraint is violated. 1 Motion Parameter Recovery The prime objective of the work is to estimate the rotation and translation parameters of a vehicle using a pair of stereo images. Corners are found in the images using the Plessey group algorithm [1]. These are matched to form a 3D point based map of the robot's environment [6]. The vehicle then moves in the ground plane. The procedure is repeated for the new coordinate frame of the AGV. A linked list of the corners qj observed at time t and their corresponding coordinates pj observed at t + 1 is formed. The motion parameters R and T are found by solution of the equation

@inproceedings{Ellwood1992GroundPM,
title={Ground Plane Motion Parameter Estimation for Non-circular Paths},
author={G. J. Ellwood and Ying Zheng and Stephen A. Billings and John E. W. Mayhew and John P. Frisby},
booktitle={BMVC},
year={1992}
}