Year: 2018
Type: dissertation
Abstract: Thanks to urban computing, it becomes possible to make the city much smarter.However, it is still difficult to understand them because such situations are affected by many complex factors, spatial-temporal dependency, and other factors.While deep learning can be deal with the complex situation through a lot of high-quality data is required, it is not sufficient in terms of spatial-temporal aspect.Even if we collect big urban data, though mixing the urban data is important, the required knowledge about it is high-level.Therefore, we propose a new framework affordable urban computing to facilitate understanding urban status by collecting and utilizing the enormous amount of urban data from limited existing urban resources.Since the city is the physical space, the texts written by people is the generic sensors to represent various status.Simultaneously, the image cameras are also very effective and very generic sensors that capture various status.With these sensors which have already existed in the city, we can exploit them to understand the city much deeper by using deep neural networks and edge devices.However, due to the nature of human activity, the distribution of the data is skewed and biased.Thus, we propose GeospaceMapping, which enables to estimate the spatial information represented in the data implicitly.In our experiment, we collected actual geotagged tweets and showed GeospaceMapping can estimate the prefecture.Then, we focus on dashboard cameras of garbage trucks which travel all around the city every day for their works film whole around the city.We propose an edge device CityInspector, which posses a function of the dashboard camera and road markings inspection and evaluate it in the experiment.Finally, we design and implement CityFlow exploiting GUI so that users can efficiently develop the application intuitively.We validate the effectiveness of CityFlow by implementing GeospaceMapping and CityInsepector.
Authors: カワノ, マコト, Kawano, Makoto, 河野, 慎
Institution: Keio University
Language: Japanese
Subfield: Urban Studies
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7139547923
