Uav-Based People Location Tracking And Analysis, For The Data-Driven Assessment Of Social Activities In Public Spaces
Jeroen Van Ameijde The Chinese University of Hong Kong
Carson, Ka Shut Leung The Chinese University of Hong Kong
Amongst an increasing awareness of the need to limit the environmental impact of urbanisation, the Compact City model offers attractive advantages such as increased land-use and transport efficiencies, walkability, and other conveniences (Dantzig, 1973, Westerink et al., 2013). The potential negative effects of high-density cities are pressures on public open spaces, micro-climates and crowding (Breheny, 1996). In high-rise cities such as Hong Kong, public spaces are a vital extension of people’s living spaces and support everyday activities (Gou et al., 2018). It is of increasing importance to study how to design inclusive, comfortable, and sociable public spaces to help increase the quality of life of urban residents. Social interactions in public spaces enable a sense of belonging, community support and help maintain health and well-being (Forrest & Wu, 2014; Lau & Murie, 2017). Social scientists have since long conducted ethnographic studies in public spaces at through manual observation techniques, which are limited in scale across time and geographical space. The research presented in this paper employs Unmanned Aerial Vehicle (UAV) mounted cameras, to capture video footage of people movements and activities in public spaces. The workflow then uses Computer Vision Object Detection (CVOD). Systems to detect and track pedestrians have been fast developing due to the vast number of possible applications such as crowd size measurement, transport security, pedestrian traffic management, detection of overcrowding situations in public buildings or tourist flow estimations (Sidla, Lypetskyy, et. al., 2006). Specialist Machine Learning processes are used to subtract human forms from background information and to discern individuals independently of the viewing angle, occlusion, or clustering in groups (shape detection). We apply the YOLOv4 algorithm that uses deep convolutional neural networks to perform object detections. The output of YOLOv4 feed these object detections into Deep SORT (Simple Online and Realtime Tracking with a Deep Association Metric) to create a highly accurate object tracker within the TensorFlow platform – an end-to-end open-source platform for machine learning. The human movements are analysed and combined into tracks, using a customised data processing workflow. Analysis from different video feeds covering different angles into the same space can be combined (Srisamosorn, Kuwahara, et. al., 2017) and integrated within digital models of the urban spaces. The aim of this methodology is to enable correlation analysis between recorded geolocation coordinates of elderly resident activities and mapped geolocation coordinates of public space facilities such as seating, exercise equipment, canopies etc. The analysis can reveal which facilities are used more often, when and for how long and how people move or interact around certain spaces. Digital visualisations of the spatial use of public spaces can engage other academics, policy makers and local residents in conversations about improvement solutions to existing public spaces, and design guidelines could be formulated to influence the creation of new public spaces as part of Hong Kong’s future urban development. References: Breheny, M. (1997). Urban compaction: Feasible and acceptable? [Elsevier Ltd]. In Cities (Vol. 14, Issue 4). https://doi.org/10.1016/s0264-2751(97)00005-x Dantzig, G. B. (1973). The ORSA New Orleans address on compact city. In Management Science (Vol. 19, Issue 10). https://doi.org/10.1287/mnsc.19.10.1151 Ghavampour, E., Del Aguila, M., & Vale, B. (2017). GIS mapping and analysis of behaviour in small urban public spaces. Area, 49(3), 349-358. Sidla, O., Lypetskyy, Y., Brandle, N., & Seer, S. (2006). Pedestrian Detection and Tracking for Counting Applications in Crowded Situations. 2006 IEEE International Conference on Video and Signal Based Surveillance, 70. Srisamosorn, V., Kuwahara, N., Yamashita, A., Ogata, T., & Ota, J. (2017). Human-tracking system using quadrotors and multiple environmental cameras for face-tracking application. International Journal of Advanced Robotic Systems, 14(5), 172988141772735. Westerink, J., Haase, D., Bauer, A., Ravetz, J., Jarrige, F., & Aalbers, C. B. E. M. (2013). Dealing with Sustainability Trade-Offs of the Compact City in Peri-Urban Planning Across European City Regions. European Planning Studies, 21(4), 473–497. https://doi.org/10.1080/09654313.2012.722927 Lau, K., & Murie, A. (2017). Residualisation and resilience: Public housing in Hong Kong. Housing Studies, 32(3), 271-295. Forrest, R., & Wu, Y. (2014). People like us? Social status, social inequality and perceptions of public rental housing. Journal of Social Policy, 43(1), 135-151. Gou, Z., Xie, X., Lu, Y., & Khoshbakht, M. (2018). Quality of Life (QoL) Survey in Hong Kong: Understanding the Importance of Housing Environment and Needs of Residents from Different Housing Sectors. International Journal of Environmental Research and Public Health, 15(2), 219.
Keywords: Public Space Studies, Human Location Tracking, Drone Scanning, Machine Learning, Sdg3: Good Health And Well-Being, Sdg11: Sustainable Cities And Communities