Predicted people per hour on every street
A neural network trained on City of Sydney walking surveys (2013–2026), four City of Sydney automatic counters (2020–2025) and Liverpool City Council people-counting cameras (2023–2024). It reads the shops, offices, food venues, stations and street types around each point from OpenStreetMap, plus hour, day type, season and weather.
Scenario: a fair-weather October 2026 day, counted as a two-way footpath cross-section, like the council surveys.
Accuracy on survey sites the model never saw: locations ranked correctly with Spearman 0.88; daily totals are off by a median 42%. Treat colours as relative busyness, and absolute numbers as estimates.
Colour ranks each street against all streets in this area. Tap any street for a catchment estimate, or a ring for measured counts.