INS - Ingénierie Numérique & Sécurité

Intelligent Retina for Innovative Sensing – IRIS

Submission summary

Intelligent vision systems are more and more important in a lot of applications nowadays, ranging from security at home, in cars, navigation of drones and robots, … but their practical applicability is still limited by the large computing system they require.

The IRIS project exhibits three facets. The first facet opens a new way by proposing new bio-inspired algorithms for the realization of a stand-alone vision chip for both optic flow (retinal slip speed) measurement and contrasting target localization over a wide luminance range. New algorithms are required because classical algorithms are efficient, but require a large amount of computation (and power) that limit their usability in such applications. Insect-based vision is not used in mainstream applications (pattern recognition, image recording) because it has comparatively lower resolution than vertebrate vision but it evolved to support accurate control of navigation in complex, 3D, dynamic environments. More than fifty years of insect research indicate that it is extremely efficient to detect motion-related events that occur when the animal moves in 3D space.

In this project, we propose to design and realize innovative algorithms adapted to the new class of 3D vision chip with 3 stacked layers: perceptive, pre-processing and computing layers.

Many studies in robotic field and insect behavior have highlighted the key role of optic flow measurement for visual navigation. Recent studies carried out at ISM in Marseille (Biorobotic department) have shown that robust optic flow measurement can be achieved by merging the output signal of a few number of insect-based motion detectors. This result combined with a technology of vertically integrated retina using 3D-stacking technology (developed by CEA), opens a promising avenue toward the implementation of fast and smart retina featuring a high fill factor, low power consumption, high flexibility and large computational resource without sacrificing the size of the overall chip. This first IRIS facet will involve ISM, LEAD, CEA-LIST and the industrial partner Novadem to determine the specifications of a new vision chip with unique features in terms of size, programmability and computational resources.

The second facet will be focused on the development of new visual processing algorithms for motion detection and object localization. These last ten years, many studies on primate visual cortex yielded novel bio-inspired models concerning rapid object categorization and recognition. The second objective of the IRIS project will aim at combining bio-inspired object categorization algorithms (extension of HMAX that will work robustly in the presence of complex scenes and developed by LEAD) with optic flow processing (developed by ISM) in order to improve the performance of the visual processing to detect and track a moving object of interest. This second facet will include a crucial benchmarking of the selected visual processing algorithms that will be implemented into the IRIS processors.

The third facet takes its place at the crossroad between the design of a novel vision chip and the use of novel control laws for the visual guidance of an aerial vehicle. Among an increasing number of flying robotic platforms, achieving collision-free vision-based piloting in cluttered indoor and outdoor environments is still a tricky challenge, even with relative large computational resources such as those offered by the IRIS sensor. Much research efforts must be deployed for the use of vision to tackle the problems involved in vital tasks and swift decisional tasks such as obstacle avoidance, odometry and path planning. It is precisely the main objective of this third facet to combine innovative visual sensors with insect-based control laws to improve the ability of future aerial vehicles (provided by Novadem) to avoid obstacles or track moving target in a natural world indoors or outdoors.

Project coordination

Stéphane Viollet (Institut des Sciences du Mouvement) – stephane.viollet@univ-amu.fr

The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.

Partner

CEA Commissariat à l'Energie Atomique et aux Energies Alternatives
LEAD Laboratoire d'Etude de l'Apprentissage et du Développement
NOVADEM NOVADEM
ISM Institut des Sciences du Mouvement

Help of the ANR 902,516 euros
Beginning and duration of the scientific project: September 2012 - 42 Months

Useful links

Explorez notre base de projets financés

 

 

ANR makes available its datasets on funded projects, click here to find more.

Sign up for the latest news:
Subscribe to our newsletter