Skip to main content

Depth and Motion Analysis

Objective

The objective of DMA was to develop a vision system integrating passive information from stereo and motion analysis for industrial robotics and passive navigation, with hardware realisation of real-time vision modules. The system will be integrated in twodemonstrators:
-a mobile vehicle to move in different environments, avoiding obstacles and making visual maps of the scene.
-a manipulating arm for industrial robots for object manipulation and inspection, and for tool assembly.
The project was organised into different tasks covering passive stereo vision, motion analysis, integration of stereo and motion, the computation and representation of 3-D shapes and motion, and hardware implementation of demonstrators in the context of amobile vehicle and a manipulating arm.
The objective was to develop a vision system integrating passive information from stereo and motion analysis for industrial robotics and passive navigation, with hardware realization of real time vision modules. The project was organized into different tasks covering passive stereo vision, motion analysis, integration of stereo and motion, the computation and representation of 3-dimensional shapes and motion, and hardware implementation of demonstrators in the context of a mobile vehicle and a manipulating arm. During the methodological period of the project, 3 strongly interrelated lines of research were followed: a study of algorithms for passive stereo and motion analysis, a hardware feasibility study, and a specification of the demonstrators (mobile vehicle and robot manipulator). In particular, a 3 camera approach was selected for stereometric purposes. A concentration on real time capabilities led to the definition and realization of a hardware architecture capable of solving stereo and motion problems at a speed suitable for application in both robot arm manipulation and vehicle guidance fields. During the realization phase a set of hardware modules was developed to implement the algorithmic chain. The project has provided systems that recognise objects well enough for a robotic inspection to be made both from mobile and stationary platforms and for the object positions to be estimated within sufficient accuracy for them to be grasped by a robot hand.
During the methodological period of the project, three strongly interrelated lines of research were followed: a study of algorithms for passive stereo and motion analysis, a hardware feasibility study, and a specification of the demonstrators (mobile vehicle and robot manipulator). In particular, a three-camera approach was selected for stereometric purposes.
Besides the need to look for a methodological solution to this class of problems, the aim of getting results as close as possible to industrial exploitation brought a remarkable concentration on real-time capabilities; this led to the definition and realisation of a hardware architecture capable of solving stereo and motion problems at a speed suitable for application in both robot-arm manipulation and vehicle guidance fields.
During the realisation phase a set of hardware modules was developed to implement the algorithmic chain. The modules are now available and their integration is at an advanced stage of progress.
DMA has provided systems that recognise objects well enough for a robotic inspection to be made both from mobile and stationary platforms and for the object positions to be estimated within sufficient accuracy for them to be grasped by a robot hand.
Exploitation
It is apparent that despite the remarkable advances achieved in understanding and implementing visual processes, vision systems are not as widely used in practice as was foreseen a few years ago.
This is mainly because of the computing flexibility required by vision processes and the unavailability of hardware machines capable of implementing them reliably in real-time.
DMA tried to find answers to these problems or, at least, to the more computation-intensive phases of a three-dimensional vision process, namely the early vision stages.
Three different kinds of demonstrations were built in the last phase of the project:
-mobile vehicle
-3-D object recognition for robot arm manipulation
-automatic 3-D model reconstruction.
All industrial partners in the project are active in markets where intelligent sensors in general, and artificial vision in particular, are considered as strategic (all these companies have been deeply involved in vision research for many years). Examples are ELSAG's applications in robotics and on measuring machines, space applications by MATRA ESPACE, and robot arm manipulation for ITMI. Besides these particular applications, the DMA architecture can be used in much more extended fields where data fusio n from multiple or time-sequence images has to be applied.

Coordinator

Elsag Bailey SpA
Address
Via Puccini 2
16154 Genova
Italy

Participants (7)

GEC-Marconi Materials Technology Ltd
United Kingdom
Address
Elstree Way
WD6 1RX Borehamwood
ITMI
France
Address
11 Chemin Des Prés
38243 Meylan
Institut National de Recherche en Informatique et en Automatique - INRIA
France
Address
Domaine De Voluceau-rocquencourt
78153 Le Chesnay
MATRA SA
France
Address
Avenue Du Centre
78052 Saint-quentin-en-yvelines
NOESIS
France
Address
27 Rue Hoche
78000 Versailles
THE CHANCELLOR, MASTERS AND SCHOLARS OF THE UNIVERSITY OF CAMBRIDGE
United Kingdom
Address
Downing Street
CB2 3EG Cambridge
Università degli Studi di Genova
Italy
Address
Via Dodecaneso 33
16146 Genova