Bio-inspired motion estimation { From modelling to evaluation, can biology be a source of inspiration?

We propose a bio-inspired approach to motion estimation based on recent neuroscience ndings concerning the motion pathway. Our goal is to identify the key biological features in order to reach a good compromise be- tween bio-inspiration and computational eciency. Here we choose the neural eld formalism which provides a sound mathematical framework to describe the model at a macroscopic scale. Within this framework we dene the cortical ac- tivity as coupled integro-dierenti al equations and we prove the well-posedness of the model. We show how our model performs on some classical computer vision videos, and we compare its behaviour against the visual system on a simple classical video used in psychophysics. Following this idea, we propose a new benchmark to evaluate models against visual system performance. Baseline results are provided for both bio-inspired and computer vision models. Results conrm the good performance of recent computer vision approaches even on such synthetic stimuli, and also show that taking biology into account in mod- els can improve performance. As a whole, this article aords a considerable insight into how biology can bring new ideas in computer vision at dierent lev- els: modelling principles, mathematical formalism and evaluation methodology. Perspectives around this work are promising and cover the addition of delays to constrain propagation as well as the extension of our benchmark to better characterise the visual system performance.