Investigation of Pedestrian-Cyclist Interactions through Machine Vision
Abstract
For pedestrian-cyclist facilities where collisions and resulting injuries may not be fully covered in police reports, there is a need for improved safety indicators. After fifteen hours of video observation at Pickard Passageway, College Station, there appears to be four broad types of pedestrian-cyclist interactions: passing, weaving, turning, and avoiding. Within each of these behavior categories, there are both safe and unsafe maneuvers. In order to determine whether an event should qualify as a safety-critical event or near-miss, multiple factors should be taken into account, including relative distance, sudden change in velocity, and sudden change in path. While an improved understanding of the general interactions between pedestrian and cyclists in these underpass facilities can lead to an improvement of the safety research field, analyzing each path manually would take a prohibitively excessive time. This paper suggests ways in which machine learning can implement the behavior categorization of pedestrian-cyclist interactions for safety evaluation at pedestrian-cyclist facilities throughout the identification, classification, and safety evaluation phases.
Citation
Gillette, George Francis (2017). Investigation of Pedestrian-Cyclist Interactions through Machine Vision. Undergraduate Research Scholars Program. Available electronically from https : / /hdl .handle .net /1969 .1 /164527.