Video Detection and Classification of Pedestrian Events at Roundabouts and Crosswalks

Project summary:

Utilization of roundabout intersections has grown substantially in the US; approximately a 10-fold increase has occurred over the last decade. Recent studies concluded significant reduction in vehicular and pedestrian crashes over their typical 4-legged intersection counterparts. They cost less to build and maintain than a signalized intersection because infrastructure to support and operate traffic signals is rarely used or required. However, other recent studies have elucidated pedestrian crossing risks. In order to address safety concerns, an appropriate diagnosis of these facilities is needed to understand if, and how, different factors contribute to crash risk. A well-established technique for studying pedestrian safety is based on reducing data from video-based in-situ observation. At present, there is no widely available, portable, 'turn-key' tool to automate extraction of these events from video recordings.

This goal of this project is to develop and test a tool based on a novel and computationally efficient image-processing algorithm that was recently developed for extraction of human activities in very complex scenes. No human intervention other than defining regions of interest for approaching vehicles and the pedestrian crossing areas will be required. The output will allow extraction basic safety performance measures such as pedestrian wait and crossing times, vehicle yield ratio, pedestrian counts, and will significantly stream-line a more detailed analysis of vehicle-pedestrian conflicts and their causal effects. The evaluation will be done using a set of multi-camera video recordings recently collected at roundabouts. Future implementations of the tool could be achieved in order to support other pedestrian safety research where extracting potential pedestrian-vehicle conflicts from video are required.