Abstract

Inadequate pedestrian and cycling infrastructure in city streets puts non-motorists across the globe at fatal risk."br" One reason for the severity of these public health and safety problems is a lack of accurate, representative data on"br" the usage of urban streets across transit modes. Unfortunately, attempts to fill this gap in understanding have been"br" limited over time by the prohibitively low reliability or high cost of current methods. Computer vision (CV)"br" systems offer a scalable and accurate platform with which to monitor traffic that overcome the sampling biases"br" imposed by most traffic measurement solutions. One such system is Numina, a standalone sensor and data platform"br" that leverages computer vision and machine learning to count and map the travel paths of street users in real time."br" Numina can concurrently detect and track travellers representing any mode of traffic and performs the majority of"br" image processing onboard each device itself, resulting in anonymized traffic data that is protective of personal"br" privacy. Using Numina and similar CV systems to enable an iterative approach to urban planning can mitigate"br" some of the risks and costs associated with legacy traffic data collection methods.