Mobile Robot Navigation in a Corridor Using Visual Odometry

Incorporation of computer vision into mobile robot localization is studied in this work. It includes the generation of localization information from raw images and its fusion with the odometric pose estimation. The technique is then implemented on a small mobile robot operating at a corridor environment. A new segmented Hough transform with an improved way of discretization is used for image line extraction. The vanishing point concept is then incorporated to classify lines as well as to estimate the orientation. A method involving the iterative elimination of the outliers is employed to find both the vanishing point and the camera position. The fusion between the vision based pose estimation and the odometry is achieved with an extended Kalman filter. A distance driven error model is used for the odometry while a simple error model with constant noise is assumed for the vision. An extended Kalman filter as a parameter estimator is also applied to estimate odometry parameters. Experimental results are included. The robustness and the precision of the entire system is illustrated by performing simple navigation tasks.