Line Based Visual Navigation Using Pose Clustering

Anders Lundquist
Computational Vision and Active Perception Laboratory (CVAP)
Royal Institute of Technology, S-100 44 Stockholm, Sweden
Licentiate Dissertation, June 1997, ISRN KTH/NA/P-97/9706-SE

Abstract

In mobile robot navigation applications it is common to have camera motion and pose (position and orientation) restricted to the ground plane. This allows the pose estimation problem to be solved using only the horizontal image position of features, using a camera aligned with the ground plane. Since these applications also often require relatively high speed performance, the 2D case of pose estimation is of potentially great importance. An analytical solution for the 2D perspective 3-point problem is given. As opposed to the 3D case, this problem has a unique solution. In order to avoid, or at least reduce, matching problems, pose clustering is used for determining the pose without explicit matching. Various ways of reducing combinatorial complexity and making use of a priori knowledge are investigated. The resulting pose estimation algorithm is then used in a Kalman filter based passive motion path estimation system as well as an active robot control navigation system.

Anders Lundquist (alun@nada.kth.se)