Objectif Research in the IMU project was directed to understanding the computational bases of vision and movement, with particular reference to scene understanding and cursive script understanding. In scene understanding (based on stereo pairs or multiple views of pictorial images as input to the system), the project focused on the interface between the early stages of computation, which provide a rich but ambiguous description in terms of low-level features, and knowledge-based processing, which generates descriptions of the 3-D organisation of the visual world compatible with the properties of the physical world. Cognitive modelling techniques based on a functional description of objects and perce ptual rules were studied. The ultimate goal was to merge the data-driven and the knowledge-driven recognitionapproaches. With regard to cursive script, basic knowledge still needs to be acquired about the mapping from linguistic material to hand trajectories; the aims of the project was to understand the writing process more than to recognise it. Techniques to code cursive script signals into symbolic descriptions were studied. Machine vision software (MAVIS) provides an image processing development environment with libraries aimed at normal and high performance requirements.Research in the project was directed to understanding the computational bases of vision and movement, with particular reference to scene understanding and cursive script understanding. In scene understanding (based on stereo paris or multiple views of pictorial images as input to the system), the project focused on the interface between the early stages of computation, which provide a rich but ambiguous description in terms of low level features, and knowledge based processing, which generates descriptions of the 3-dimensional organization of the visual world compatible with the properties of the physical world. Cognitive modelling techniques based on a functional description of objects and perceptual rules were studied. The ultimate goal was to merge the data driven and the knowledge driven recognition approaches. With regard to cursive script, basic knowledge still needs to be acquired about the mapping from linguistic material to hand trajectories; the aim of the project was to understand the writing process more than to recognise it. Techniques to code cursive script signals into symbolic descriptions were studied. Basic libraries of computational modules that perform early processing of image data (including the integration of stereo and motion algorithms to obtain depth information) and cursive script data were developed. A portable interactive software environment was developed. This makes possible the generation and interrogation of multiple regional representations of images or image sequences (iconic representations, regional representations, contour representations). Cognitive driven modules that interface with low level representations of images using either basic perceptual rules for dealing with occlusions of surfaces in low level representations or semantic representations of objects for interpreting perceptual representations and describing scenes have been developed. The firmware necessary for the real time acquisition of sequences of stereo pairs o f images has been developed.The following results were obtained: -Development of basic libraries of computational modules that perform early processing of image data (including the integration of stereo and motion algorithms to obtain depth information) and cursive script data. -Development of a portable interactive software environment, VIS. This makes possible the generation and interrogation of multiple regional representations of images or image sequences (iconic representations, regional representations, contour representa tions). It has been implemented on PC-AT, Transputer network, 68020, VAX, etc. -Development of cognitive-driven modules that interface with low-level representations of images: .using basic perceptual rules for dealing with occlusions of surfaces in low level representations; .using semantic representations of objects for interpreting perceptual representations and describing scenes. -Development of the firmware necessary for the real-time acquisition of sequences of stereo pairs of images on the VDS Eidobrain workstation and the porting of VIS on the Eidobrain. Exploitation The results of the project will serve the R&D community, both academic and industrial, as conceptual and computational support when developing specific applications of image or movement analysis. Moreover, the VIS system, incorporating well-engineered implementations of advanced algorithms, could form the basis for a marketable product. Champ scientifique natural sciencescomputer and information sciencessoftwarenatural sciencescomputer and information sciencesartificial intelligencecomputer vision Programme(s) FP1-ESPRIT 1 - European programme (EEC) for research and development in information technologies (ESPRIT), 1984-1988 Thème(s) Data not available Appel à propositions Data not available Régime de financement Data not available Coordinateur Università degli Studi di Genova Contribution de l’UE Aucune donnée Adresse Via all'Opera Pia 11A 16145 Genova Italie Voir sur la carte Coût total Aucune donnée Participants (4) Trier par ordre alphabétique Trier par contribution de l’UE Tout développer Tout réduire CAPTEC-COMPUTER APPLIED TECHNICS Irlande Contribution de l’UE Aucune donnée Adresse ST.JAMES TERRACE X MALAHIDE Voir sur la carte Coût total Aucune donnée KATHOLIEKE UNIVERSITEIT NIJMEGEN Pays-Bas Contribution de l’UE Aucune donnée Adresse MONTESSORILAAN 9104 6500 HE NIJMEGEN Voir sur la carte Coût total Aucune donnée TRINITY COLLEGE DUBLIN Irlande Contribution de l’UE Aucune donnée Adresse X DUBLIN 2 Voir sur la carte Coût total Aucune donnée VIDEO DISPLAY SYSTEMS SRL Italie Contribution de l’UE Aucune donnée Adresse VIA MUZIO CLEMENTI 16 50127 FIRENZE Voir sur la carte Coût total Aucune donnée