01 JUN 2021 (TUE) 10:05-10:25

A geo-intelligent approach to analyze walkability


Zoom Link: https://hku.zoom.us/j/97166843596

Meeting ID: 971 6684 3596

Password: 016676

Moderator: Dr Steven H.S. Zhang


Ms LIAN Ting PhD Student, Department of Geography, HKU

Supervisor:

Professor Becky P Y Loo


Abstract:

Walking, one of the most common physical activities, has proven to have positive effects on human physical and mental health. However, rapid urbanization and motorization can compress and degrade walking space, which jeopardizes the walking experience. Assessing walkability systematically and scientifically is necessary for planning a pedestrian-friendly transport system. Although many studies have evaluated walkability by considering various indicators, the focus is primarily on static and place-centered indicators (e.g., density of facilities, distance to facilities, land use mix, street greenery, etc.). The behaviors and movements of pedestrians are not sufficiently addressed. This thesis aims to enrich walkability assessments from a place-based and people-oriented approach by considering dynamic and social indicators related to pedestrian activities and street life. With the help of advanced computing power and Artificial Intelligence (AI) algorithms, this study will capture and estimate dynamic pedestrian-centered factors (i.e., pedestrian volume, pedestrian density) from bus dashcam videos by using deep learning methods, such as Fast Region-based Convolutional Network (Fast R-CNN) and Fully Convolutional Networks (FCN). Supplemented by static features derived from street view images and other geospatial datasets (i.e., road network data, POI data, etc.), this study will analyze walkability in its three fundamental dimensions of pedestrian safety, convenience, and comfort. The respective key indicators are pedestrian exposure (to the risk of pedestrian-vehicle collisions), pedestrian detour ratio, and pedestrian crowding, respectively. Geo-intelligent approaches which take advantage of deep learning methods and dynamic big data embedded in bus dashcam videos will be used. Random forest regression will be used to build predictive models to predict the values of these indicators in areas not covered by bus dashcam data. Finally, the three indicators will also be integrated into a holistic assessment of walkability with other indicators. The methods and the workflows can provide technical support for implementing smart city governance and building pedestrian-friendly neighborhoods.