GRIDSMART does not miss bicyclists on the imagery it has been tested against. The GRIDSMART algorithm has been trained on and tested against more than a million hand-annotated images gathered from dozens of intersections in various environments around the world. When tested on annotated unseen data (i.e., images not in the training data), GRIDSMART achieves greater than 99% bicycle discrimination accuracy on a single image frame. Each bicyclist, however, is evaluated multiple times, usually 10 or more, when traveling through the intersection, thereby providing multiple opportunities to detect the tracked object as a bicyclist, providing unparalleled accuracy, safety, and efficiency.
Note, however, that all machine learning algorithms, particularly deep neural networks, are only as good as the data with which they were trained. It is possible that some intersections will present new conditions that should be added to GRIDSMART’s suite of training data to improve performance. The GRIDSMART FAE organization will work closely with you in that unlikely scenario so that GRIDSMART can continue to improve.