Ground truth data
Part of Thesis summary. The study area, the map layers, and how a field survey produced 907 stays with known correct answers.
- Study area: Hiroshima City, and the MILES GPS dataset the method is built for
- Spatial entity layers: how the map was split into buildings, streets and open space
- Experiment design: the field survey that produced the correct answers
- the five groups of stops: Group A: vertical position, Group B: sky view obstruction, Group C: indoor position, Group D: movement type, Group E: ambiguous edge cases
- Self-report form: the form participants filled in at each stop
- Downsampling to MILES density: removing survey points so the data looks like MILES
- Ground truth cleaning: from 1,029 records to 975, then to 907
- Ground truth drop list: what was removed in the last cleaning step
- Ground truth dataset: the final 907 stays
Meetings: RQ2 diagnostics and RQ3 ceiling check (975 vs 956 count), Does downsampling matter? (does downsampling matter?).