日本語のREADMEはこちらです: README.ja.md
This repository publishes aggregated open data from AI cameras installed at key tourist locations in Fukui Prefecture, Japan. Data is regularly collected, processed, and released for public use and visualization.
A web application is available to visualize the people flow data.
AI cameras are installed at the following key tourist locations in Fukui Prefecture:
- Fukui Station East Entrance
- Tojinbo Cliffs
- Rainbow Line Summit Park (Parking Lot 1 & 2)
full/: Complete aggregated CSV data, separated by location and detection target.monthly/: Data aggregated by day, organized bylocation/target/year/month.daily/: Data aggregated by hour, organized bylocation/target/year/month/day.hourly/: Data at 5-minute intervals, organized bylocation/target/year/month/day/hour.tools/: TypeScript scripts for data aggregation and processing (run with Deno).
Scripts in the tools/ directory (run with Deno):
aggregate-day.deno.ts,aggregate-hour.deno.ts,aggregate5mins.deno.ts: Aggregate raw CSV data into daily, hourly, and 5-minute intervals.csv2sqlite.deno.ts: Convert CSV data to SQLite databases.check-csv.deno.ts: Validate CSV files for errors or warnings.escape-age-data.deno.ts,escape-movement-data.deno.ts: Fix formatting issues in age and movement data.license-plate-aggregation.deno.ts: Aggregate license plate data from parking lots.thin-out-to-one-second-interval.deno.ts: Reduce movement data to one-second intervals.
CSV files contain aggregated counts and attributes for three types of detected objects:
Person: General person detection.Face: Face detection with estimated attributes like age and gender.LicensePlate: Vehicle license plate detection with estimated prefecture of origin and vehicle type.
- Camera Hours: All cameras operate 24/7, except during maintenance or equipment failures.
- Zero Counts: A value of
0in the data indicates that no objects were detected within the camera's field of view during that interval. It does not necessarily mean the camera was offline.
- Face Detection: Operates 24/7 but with significantly lower detection counts during early morning (before 8 AM) and late evening (after 7 PM). This is due to several factors:
- Outdoor lighting conditions greatly affect facial recognition accuracy.
- At Fukui Station East Entrance, security shutters are lowered outside business hours, limiting camera field of view.
- The detection algorithm naturally has lower sensitivity in low-light conditions.
- Actual visitor volumes are much lower during these periods.
- Despite lower counts, the camera is still recording and processing faces.
- Person Detection: Available 24/7 with consistent detection capability across all hours. More reliable than Face detection for complete 24-hour analysis.
- License Plate Detection: Available 24/7 when the camera system is operational.
- All detection types operate continuously 24/7 (except during maintenance or known outages).
- Face Detection shows natural variation throughout the day:
- Very low detections: 12:00 AM - 8:00 AM (averaging <100 detections/hour at Fukui Station).
- Peak detections: 11:00 AM - 5:00 PM (averaging >12,000 detections/hour at Fukui Station).
- Lower detections: 7:00 PM - 11:00 PM (averaging <500 detections/hour).
- Person Detection shows more consistent 24-hour coverage with reasonable detection counts across all hours.
- The data reflects actual visitor patterns with peak activity during business hours and tourist season.
- Hourly data is provided at 5-minute intervals for detailed temporal analysis.
- Data Collection Start: Data collection began on December 20, 2024. No data is available prior to this date.
- System Outage: A complete system outage occurred at all locations from September 26-28, 2025.
- Equipment Failure: The license plate detection camera at Rainbow Line Parking Lot 1 experienced operational issues (specific dates pending documentation).
This project is licensed under the terms of the MIT license.