5th International Conference on Simulation of Adaptive Behavior, Zürich, Switzerland, 17 - 21 August 1998, pp.114-119
This paper describes and evaluates visual looming as a method for monocular range estimation. The looming algorithm is based on the relationship between displacements of the observer relative to an object, and the resulting change in the size of the object's image on the focal plane of the camera. Though the looming algorithm has been described in detail in prior reports, its usefulness for inexpensive, robust ranging has not been realized widely. In this paper we analyze visual looming as a visual range sensor for autonomous mobile robots. Systematic experiments with a Pioneer I mobile robot show that visual looming can be used to extract ranging information much as with sonar. The accuracy of the looming algorithm is found to be significantly more robust than sonar when the object whose distance is being measured is slanted relative to the robot's line of sight. On the other hand, sonar is better suited for objects that cannot be visually segmented from their background, or objects that do not fit entirely within the focal plane. The results suggest that looming can be used as a robust, inexpensive range sensor as a complement to sonar.