Interactive Project · Street Equity & Inclusiveness

i

EgoCity

Who gets to use the street? · Haidian, Beijing

What if a city could see itself through the eyes of the people who move through it all day? EgoCity reads Beijing’s streets from cameras carried by food-delivery riders — turning 220 hours of everyday rides into a fine-grained portrait of who actually uses public space, and who is left out.

Scroll to explore ↓

8delivery riders
220+hours of video
118,240street frames
91,524pedestrians read
2,106street grids
96%model F1 accuracy

The question

A street can be busy and still be unfair

Streets are the most ubiquitous public space we have — yet we rarely measure whether they actually serve everyone. EgoCity asks two things of every block: is it fully and diversely used (utilization), and does it serve all age groups fairly (inclusiveness)? A vibrant street that only adults occupy is busy, but not just.

Interactive · In progress

Rider trajectories across Haidian

The animated EgoCity routes — eight riders sweeping the district over three weeks — will play here.

The method

The street, seen from a scooter

Stage 1 · Collect

EgoCity dataset

Eight food-delivery riders carried portable cameras across Haidian over three weeks — yielding 220+ hours of video and 118,240 anonymized street frames from a pedestrian’s-eye view.

Stage 2 · Identify

LLaVA-PI (a VLM)

A hybrid vision-language model detects each person and labels them child / adult / elderly plus six activities — at 0.96 F1, outperforming GPT-4o and Gemini-1.5-Pro.

Stage 3 · Evaluate

Two indices

On 100 m grids, we measure Utilization (volume × activity diversity) and Inclusiveness (how evenly age groups appear, relative to who lives there).

Individual level

Who’s actually on the street?

Of 91,524 people read from the frames, the street skews overwhelmingly adult. Set against who actually lives in Haidian, the gap is stark: the elderly are nearly one in five residents but barely one in forty people on the street.

On the street
Residents
Adults Elderly Children

What people do

Mostly passing through

Across the six recognized activities, walking dominates and lingering is common; talking is occasional and phone-calling almost absent. Bars show the share of street grids where each activity appears. (Mobile browsing and dog-walking sit between talking and calling.)

Walking
89%
Staying
55%
Talking
20%
Phone calling
2%

What we found

Four readings of the street

Adults dominate

96% of detected pedestrians were adults; children and the elderly were strikingly underrepresented relative to their share of residents.

Presence ≠ population

Pedestrian flow and residential population were spatially mismatched — dense residential streets stayed quiet, while moderate-density areas like Zhongguancun buzzed.

Vitality concentrates

High utilization clustered in a few cores (Zhongguancun–Haidian, Balizhuang), while peripheries such as Ganjiakou and Lugouqiao stayed persistently low.

Vibrant ≠ equitable

Very few grids were both highly used and highly inclusive — the liveliest streets were often the least balanced across age groups.

Why it matters

Cheap data, fairer streets

By riding along with workers already moving through the city, EgoCity captures fine-grained, pedestrian-eye data at a fraction of the usual cost — and gives planners a way to spot streets that are busy but exclusionary, and to revive under-used residential blocks, toward streets that hold children and elders, not just adults.

Project

EgoCity

EgoCity reads Beijing’s public streets from the cameras of food-delivery riders. Pairing a crowdsourced, pedestrian-eye video dataset with a custom vision-language model, it measures who actually uses the street — and how fairly space is shared across children, adults, and the elderly.

Team

Mingze Chen — Research Supervision
Xiamengwei Zhang — Research Lead
Mingze Chen — Visualization & Web

A Nature AI Lab project.

Publication

Zhang, X., Chen, M., Huang, Y. (2025). Who gets to use the street? Evaluate the utilization and inclusiveness using crowdsourced videos and vision-language models. Sustainable Cities and Society.

Use & Credit

The visualizations on this page may be reused in any publication provided that:

  1. They are duly credited as a project by the Nature AI Lab;
  2. A copy of the publication is sent to mingze.chen@ubc.ca.

For more information, mingze.chen@ubc.ca

NATURE AI LAB natureailab.com
Lab Publications ↗ Lab News ↗