This paper describes procedures for obtaining a reliable and dense optical flow from image sequences taken by a TV camera mounted on a car moving in usual outdoor scenarios. By using correlation based techniques and by correcting the optical flows for shocks and vibrations, useful sequences of optical flows can be obtained. When the car is moving along a flat road and the optical axis of the TV camera is parallel to the ground, the motion field is expected to be almost quadratic and have a specific structure. As a consequence the egomotion can be estimated from this optical flow and information on the speed and the angular velocity of the moving vehicle are obtained. By analyzing the optical flow it is possible to recover also a coarse segmentation of the flow, in which objects moving with a different speed are identified. By combining information from intensity edges a better localization of motion boundaries are obtained. These results suggest that the optical flow can be successfully used by a vision system for assisting a driver in a vehicle moving in usual streets and motorways.

The robot moves in a simplified world and performs tasks of increasing complexity, such as the exploration of a maze, the recovery of its structure and the planning of some nearly optimal trajectories through the maze. There are three layers of robot motion control: the lowest layer is essentially an optomotor reflex; the second layer has a short term memory and operates when the visual information is not complete; the third layer has a long term memory and reasoning capabilities. The vision and motor systems are assumed to be finite state automata and the control architecture is based on transitions between states of these automata. The intelligent behavior of the robot is primarily caused by the memory and the capability to create simple and useful representations of the world.>

This paper describes some experiments on control strategies of a mobile robot. The robot has a simple vision system and moves in a simplified world, but the controlling architecture performs tasks of increasing complexity, such as the exploration of a maze, the recovery of its structure and the planning of some nearly optimal trajectories through the maze. There are three layers of control of robot motion: the lowest layer analyses images, controls the speed of robot wheels and is essentially an optomotor reflex; the second layer has a short term memory and operates when the visual information is not complete or when the robot has lost its way; the third layer has a long term memory and reasoning capabilities, monitors the achievement of robot tasks and schedules the timing of completion of the different tasks. The vision and motor systems and other supervising units are assumed to be finite state automata and the control architecture is based on transitions between states of these automata. The control system is flexible and new capabilities are easily acquired by simply introducing new states. The intelligent behaviour of this artificial system is primarily caused by the memory and the capability to create simple and useful representations of the world.>

Insects and a lot of other animals use the optical flow to control the direction of their motion and to avoid obstacles. This paper describes experiments suggesting the possible use of the optical flow for the navigation of a robot moving in indoor and outdoor environments. In indoor scenes, such as corridors, offices and laboratories, the optical flow is used to detect and localize obstacles. These routines are based on the computation of a reduced optical flow. Almost real time performance was obtained with standard workstations, such as SUN 3 or SUN Sparcstation 1. The mobile vehicle is usually able to avoid large obstacles such as a chair or a human, but it is not able to avoid thin obstacles such as a rod or a bar. The avoidance performances of the proposed algorithm critically depend on the feedback loop between the vision module and the motor system. In outdoor scenes the optical flow can be used to understand the egomotion, that is to obtain information on the absolute velocity of the moving vehicle. The optical flow is corrected for shocks and vibration present during image acquisition. Regions of the image are extracted, where the optical flow is reliable, and the information on egomotion is recovered from the optical flow here obtained. These results suggest that the optical flow can be successfully used by biological and artificial systems for controlling their motion and for avoiding obstacles.

A general procedure for the understanding of a class of indoor scenes is proposed. The suggested algorithm makes use of an extensive 2-D processing of the image that provides a faithful line drawing of the viewed scene. It also uses procedures adequate for the understanding of the 3-D structure and for the recognition of several items in the scene. The approach is specific for indoor scenes, such as corridors, offices, and laboratory rooms, where long straight edges are predominant and vanishing points can be detected.>

The main goals of Computer Vision are the recovery of the three dimensional structure and the recognition of viewed objects from T.V. images. The understanding of images of indoor scenes, such as corridors and offices, is greatly simplified by the exploration of geometrical properties of the viewed scene. In these scenes there are many long straight edges which are either parallel or orthogonal and the viewed objects can be modeled as elements of Legoland. Legoland is a block world where physical edges have at most three directions, which are mutually orthogonal. In this paper a general procedure is proposed for the understanding of indoor scenes which is likely to be useful in Computer Vision. The proposed algorithm makes use of an extensive 2D image processing which is able to provide a faithful line drawing of the viewed scene. It also uses procedures adequate for both the understanding of the 3D structure of the viewed scene and for the recognition of several items in the scene.

A method is presented for the recovery of optical flow. The key idea is that the local spatial structure of optical flow, with the exception of surface boundaries, is usually rather coherent and can thus be appropriately approximated by a linear vector field. According to the proposed method, the optical flow components and their first order spatial derivatives are computed at the central points of rather large and overlapping patches which cover the image plane as the solution to a highly overconstrained system of linear algebraic equations. The equations, which are solved through the use of standard least mean square techniques, are derived from the assumptions that the changing image brightness is stationary everywhere over time and that optical flow is, locally, a linear vector field. The method has been tested on many sequences of synthetic and real images and the obtained optical flow has been used to estimate three-dimensional motion parameters with very good results.>