Streetscope: Measuring Street-level Conditions From Wearable Camera Transects

Analyzing gender representation and infrastructure quality in Mumbai and Navi Mumbai using GoPro wearable camera footage.

Key Findings

Proportion Women

City	Images	Unweighted	Weighted	Ped. Sex Ratio
Mumbai	2,740	0.147	0.160	238.8
Navi Mumbai	389	0.131	0.131	219.7

Women are significantly underrepresented in public spaces, comprising only 13-16% of visible people. The pedestrian sex ratio (females per 1000 males) is far below census baselines (Mumbai: 838, Navi Mumbai: 910).

By Mode

City	Pedestrians (unweighted)	Pedestrians (weighted)	Two-wheelers (unweighted)	Two-wheelers (weighted)
Mumbai	0.185	0.193	0.071	0.083
Navi Mumbai	0.186	0.180	0.045	0.050

Proportion of women is consistently lower among two-wheeler riders than pedestrians.

By Road Type

City	Primary	Secondary	Tertiary	Residential
Mumbai (prop women)	0.164	0.127	0.164	0.162
Mumbai (sex ratio)	196	146	197	193
Navi Mumbai (prop women)	0.090	0.110	0.145	0.179
Navi Mumbai (sex ratio)	99	123	169	219

In Navi Mumbai, there is a clear gradient: proportion of women increases from primary roads (9%) to residential roads (18%). Mumbai shows less variation across road types.

Temporal Patterns

Data coverage: ~7 AM - 7 PM IST. Thursday overrepresented (40%).

Weekday vs Weekend:

City	Weekday	Weekend	Significant?
Mumbai	0.163 (n=2100)	0.145 (n=458)	Yes (p=0.048)
Navi Mumbai	0.138 (n=835)	0.168 (n=129)	No

Mumbai shows significantly higher proportion of women on weekdays. Navi Mumbai shows the opposite pattern but the difference is not statistically significant.

Data Collection by Time Period (equal 4-hour windows):

Time Period (IST)	Mumbai	Navi Mumbai
Morning (7-11)	0.21	0.39
Midday (11-15)	0.41	0.47
Evening (15-19)	0.37	0.13

Columns sum to 1.0 within each city. Navi Mumbai data is concentrated in morning/midday hours with sparse evening coverage.

See scripts/analysis/temporal_analysis.ipynb for detailed temporal analysis including distribution plots and POI × time interactions.

By Points of Interest (POI)

POI	Yes	No	Significant?
Bus Station	0.168 (n=137)	0.156 (n=3385)	Yes (p=0.04)
Railway Station	0.141 (n=7)	0.156 (n=3515)	No
Street Vendor	0.208 (n=238)	0.150 (n=3283)	Yes (p<0.001)

Street vendors are associated with significantly higher proportion of women. Bus stations show a small positive effect.

Infrastructure

Footpath:

City	Paved	Paved-Blocked	No Sidewalk	N
Mumbai	1,150 (59%)	324 (17%)	459 (24%)	1,938
Navi Mumbai	217 (74%)	17 (6%)	60 (20%)	294

Litter:

City	Yes	Construction Debris	No	N
Mumbai	256 (78%)	57 (17%)	16 (5%)	329
Navi Mumbai	54 (93%)	3 (5%)	1 (2%)	58

Lane Markings:

City	Yes	No	N
Mumbai	1,602 (99%)	12 (1%)	1,614
Navi Mumbai	318 (100%)	0	318

Potholes:

City	Yes	No	N
Mumbai	30 (54%)	26 (46%)	56
Navi Mumbai	3 (75%)	1 (25%)	4

Note: Infrastructure annotations are sparse - not all images have all fields.

Data Pipeline

GoPro Videos
     |
scripts/extract_exif.py         Extract EXIF metadata from videos
     |
scripts/extract_gps_timeseries.py   Parse GPS data from EXIF files
     |
scripts/extract_frames.py       Extract frames at regular intervals
     |
scripts/compress_frames.py      Compress frames for annotation
     |
Label Studio                    Human annotation of images
     |
scripts/analysis/*.ipynb        Analysis and visualization

Setup

System Tools

# macOS
brew install ffmpeg exiftool

# Ubuntu
sudo apt-get install ffmpeg exiftool

Python Packages

pip install pandas Pillow tqdm
pip install opencv-python  # optional fallback for video info

Usage

# 1. Extract EXIF metadata from videos
python scripts/extract_exif.py \
    --input /path/to/videos \
    --exif-output exif/ \
    --metadata-csv video_metadata.csv

# 2. Extract GPS timeseries from EXIF files
python scripts/extract_gps_timeseries.py \
    --input exif/ \
    --output gps_timeseries.csv

# 3. Extract frames every 10 seconds
python scripts/extract_frames.py \
    --input /path/to/videos \
    --output frames/ \
    --log frame_log.txt \
    --report extraction_report.csv \
    --every-seconds 10

# 4. Compress frames for annotation
python scripts/compress_frames.py \
    --input frames/ \
    --output annotation_frames/ \
    --resolution 1280x720 \
    --quality 75

# 5. Run analysis notebooks
jupyter notebook scripts/analysis/mumbai_annotations.ipynb
jupyter notebook scripts/analysis/navi_mumbai_annotations.ipynb

Data

• Mumbai: 2,740 annotated images from GoPro transects
• Navi Mumbai: 389 annotated images from GoPro transects
• Annotation fields: men_count, women_count, men_twowheeler, women_twowheeler, footpath, lane_markings, potholes, litter, bus_station, railway_station, street_vendor

Directory Structure

annotation_frames/      Compressed frames for Label Studio
data/                   Data files (CSVs)
exif/                   EXIF metadata files
frames/                 Extracted video frames
labelstudio/            Label Studio JSON exports
output/maps/            HTML visualizations (maps, heatmaps)
scripts/
  analysis/             Jupyter notebooks for analysis
  extract_exif.py       Extract EXIF from videos
  extract_gps_timeseries.py   Parse GPS from EXIF
  extract_frames.py     Extract frames from videos
  compress_frames.py    Compress frames for annotation

Output Files

File	Description
output/maps/mumbai_annotations_map.html	Interactive map of Mumbai annotations
output/maps/navi_mumbai_annotations_map.html	Interactive map of Navi Mumbai annotations
output/maps/mumbai_sex_ratio_heatmap.html	Mumbai sex ratio heatmap
output/maps/navi_mumbai_sex_ratio_heatmap.html	Navi Mumbai sex ratio heatmap
output/maps/mumbai_heatmap.html	GPS point density heatmap
output/maps/mumbai_route_map.html	Route visualization
data/mumbai_annotations_with_exif.csv	Merged annotation and GPS data

License

MIT