From a scientific/technological point of view, we expect that the impact resulting from the holistic approach for the design and implementation of vision systems proposed in this project will revolve around the following aspects:
i) Setting a global context for the design of vision hardware components, reaching even low-level structures ―ultimately full-custom circuits on a chip―, will provide practitioners with guidelines about how to optimally partition and distribute resources. Typically, system designers work upon datasheets of components independently characterized. Drawing conclusions about the impact of certain parameters on the rest of system elements requires bechmarking, modeling and careful assessment of critical metrics for the targeted application scenario. A global exploration context will enable the propagation of advantages and limitations for various alternative realizations across the complete sensing-processing hierarchy. This methodology is actually connatural to other industrial sectors dealing with complex systems, e. g. aerospace, where multi-level optimization across different domains is mandatory.
ii) As a direct consequence of such multi-level optimization, the techniques to be developed on energy harvesting, early acceleration of low-level tasks, 3D sensing, memory allocation etc. will be evaluated beyond raw standard measurements. Their performance analysis will take into account aspects like affection of the algorithm dataflow, possible generation of bottlenecks, integration within the hardware-software framework, effective interfacing, implications at application level etc. This will enable further consideration of those techniques and corresponding components from a vertical perspective, not constrained to the usual horizontal characterization.
iii) We will work thoroughly on demo prototypes that will allow us to increase the visibility of our groups in different scientific communities and industrial environments dedicated to circuit design, embedded vision, algorithms, system design etc. We employ this impulse to disseminate, contrast and boost our results in order to keep being a reference in our field.
The expected socio-economic impact is defined in the following terms:
i) Industrial property protection. The members of the consortium are holders of various industrial property titles, including national and foreign patents. Some of these patents have been licensed, which demonstrates the team's willingness and capability to protect research results suitable for commercial exploitation.
ii) Market of artificial vision. It is expected that this market will multiply its volume by a factor 25 in the next 7 years. Giving these circumstances, the results achieved in this project will definitely attract interest not only from academia but also from industry. In particular, we will do our best to engage companies related to transportation as the Society Challenge primarily addressed by this project.
iii) Market of image sensors. It is also expected that this market will experience a growth over 10% in the next few years. Key features like depth sensing and always-on operation will be critical to disrupt new niches. The incorporation of near-sensor intelligence is also gaining relevance for image sensor designers and manufacturers, with a focus on automotive applications. Therefore, we will try to engage companies in this field as well. In particular, we will promote the advantages of the proposed holistic approach through prototype smart image sensors designed to boost the performance of vision algorithms rather than to exclusively maximize the image quality and frame rate.