Kinematic parameters estimation of robot motion using active vision

The authors address the problem of estimating the distance and velocity, between a mobile robot and an obstacle. The goal is to determine the kinematic parameters of the relative motion to build an effective avoidance motion. The information issued from a camera coupled with a laser stripe is used to estimate the kinematic parameters with an extended Kalman filter (EKF). The relative motion between the vehicle and the obstacles is not supposed to be known, so the state evolution equations are unknown. The evolution model for the EKF is presented as a movement with constant acceleration. This model is a local approximation of a real movement by a quadric. As this model will often be a rough approximation of the reality, the filter will certainly diverge for almost all movements. Simulated results are given, the problem of the filter divergence is discussed, and a solution is proposed. The real implementation on a transputer network is described