NEXYAD is starting a new image capture campaign in the summer heat of Paris suburb. Those new captured videos will sample the issue of strong casted shadows and will show robustness of our two Artificial Intelligence perception algorithms RoadNex (road border detection even without markings and free space detection) and ObstaNex (obstacles detection).
Database of NEXYAD applies the methodology A.G.E.N.D.A. (Approche Générale des Etudes Neuronales pour le Développement d’Applications) that allows to train deep learning with a compact learning database (knowledge-based selection of learning and of validation examples) with much better performance than with randomly selected massive databases.
Yes it is possible ! …
NEXYAD can then re-train ANNs for a special cam for instance, very quickly, with real maths KPIs instead of poor « percentages » estimators of effectiveness and robustness.
Try our AI perception algorithms, it will change a few things :
. VERY robust detection
. low computer load consumption (runs on regular architectures)
. retrainable quickly is sensor technology evolves
