Counting Bicycles

Airel view of intersection
Engineering

C. Alec Gosse
Civil and Environmental Engineering

Emmanuel Denloye-Ito
Electrical and Computer Engineering

Motivation: Realizing the social and environmental benefits of bicycle transportation requires good planning. Good planning requires continuous observation of ridership patterns, which are difficult to collect.

Opportunity: Modern traffic signals use video to detect waiting vehicles. Enhancing this capability to determine ground trajectory and
vehicle type will determine turning movement counts continuously without additional sensor costs. Method: Superpixels identify moving objects in image space and
enable tracking across individual camera views. Examining vehicle perspective shift allows estimation of physical attributes and accurate ground trajectories. This reduces massive video data to compact vehicle "signatures". Probabilistic matching of signatures between camera views then recognizes turning movements despite incomplete information.  

Implications: Low-cost continuous count data will allow inherently variable bicycle travel to be understood and accommodated to the same standard as automobile travel.