A Computer Vision Framework for Human User Sensing in Public Open Spaces.

The field of Urban design considers how people utilize public open spaces (POS) when designing spaces such as parks, plazas, and streets. Current methods of observing public space use rely on visual observation which consumes much time and effort to detect users' physical activities in large POS; these methods also only provide qualitative observations of how patrons behave in these areas. Active sensors, such as wearable sensors and smart phones with GPS tracking capabilities, have high costs and cannot sense all users in a POS (namely, such sensors are "blind to" those without wearable sensors). Therefore, it is appealing to make use of video data from pre-installed surveillance cameras in POS to extract POS use information from video using computer vision methods. This paper proposes a sensing framework based on computer vision to measure human activities in POS. As part of the study, an extensively labeled datset of people and their activities in POS (termed OPOS) is used to train detectors. A case study of the proposed framework is presented using security camera feeds from a greenway at the Detroit Riverfront. The AP0.50 results of the trained detector are 96.3% for pedestrian detection and 96.5% for cyclist detection, respectively. These results show such an approach can reliably track patrons in parks to ascertain their behavior and to inform future POS improvements.