Deep geographic embodiment

We examine embodiment as a geographical concept and construct using novel spatial data collection and deep learning on streetscapes. Embodiment evokes the varied geographies of how people, going about their lived experiences, mentally build and refine geographic information in fleeting bouts of encounter with the physical substrate and social milieu of the moment. We show that this information can be examined, recorded, and organized empirically through a combination of mobile ethnography, wearable computing, and deep learning. Mobile ethnography and wearable computing, we argue, constitute a new form of realistic, immersive, and context-rich spatial data collection, which we term “auto-glancing”. Deep learning can transform these data into new forms of geographic information, producing what we call “deep geographic embodiment”. We show an example of this deep geographic learning as fused object detection, instance segmentation, multiple-object tracking, pose detection, surface mapping, and real-time action recognition. We will argue that auto-glancing and deep geographic embodiment have the potential to produce geographic information systems with one-to-one mappings to reality. This new information geography could bring conceptual debates in human geography into alignment and parity of explanation with machine learning and knowledge discovery from computer science, with mutually-reinforcing benefits. • We address how geographic information may reveal momentary geographies of life, building empirical support for embodiment questions in spatial behavior. • We introduce “auto-glancing” as spatial data science for embodiment, based on real-time video feeds. • We introduce a deep learning scheme, “deep geographic embodiment” that can convert auto-glanced data into information geography. • We demonstrate worked examples on ∼1,300 fixed site observations and 52.272 million video frames of everyday pedestrian encounter in New York City. • We show deep geographic embodiment advancing the geographical sciences, re-energizing empirical human geography and tackling the modifiable areal unit problem.