This dashboard supports cycling-behaviour-informed policy planning for urban planners and officials.
It combines cycling sensor data (speed and road quality) with aggregated road segment analysis, fetched directly from the cloud.
The map visualises the local cycling network using selectable data layers. Each layer has its own scale, legend, and policy interpretation.
Displays average cycling speed across recorded trips, supporting the identification of slow or fast corridors.
Based on third-party infrastructure data, categorised from "Perfect" to "No road", supporting infrastructure assessment and maintenance planning.
Aggregates speed and road quality into averaged scores per road segment, supporting corridor-level analysis.
Within the averaged road segments, you'll see a subfilter which combines infrastructure condition and cycling performance into a single indicator for comparative assessment.
Flags locations where a cyclist decelerated suddenly, accumulated across all recorded trips. Braking is detected when a cyclist decelerates at more than 2 km/h per second— meaning speed drops by at least 2 km/h within a single one-second GPS interval. This threshold was adjusted so as to distinguish genuine sudden braking (e.g. at intersections, obstacles, or unsafe road conditions) from normal slowdowns.
A "g" is a unit of acceleration equal to Earth's gravity (9.8 m/s²)— the standard way to measure impact force.
An incident is flagged when the accelerometer registers a sharp impact of 6g or greater. Because potholes, rough terrain, and other everyday jolts can also produce a brief spike above that threshold, a raw impact reading isn't treated as a confirmed crash until it passes three independent checks:
1. The bike actually stops. Within 3 seconds of the impact, the sensor readings must settle into a narrow band (varying by no more than 3g) for at least 1 continuous second, and GPS must confirm the bike stopped moving— its reported speed dropping near zero and its position staying put— for that same stretch. Both signals have to agree; if either keeps showing motion or vibration, the spike is treated as normal riding activity (e.g. hitting a pothole and continuing to ride).
2. The wheel actually stalls. Separately from the accelerometer/GPS check above, the system watches the wheel's own rotation for a stall— the point where it stops turning and, if the bike gets back underway, resumes. If that recovery takes longer than 120 seconds, the event is treated as a parking artifact (e.g. the bike being walked, lifted, or left standing) rather than a crash, and is rejected even if the first check passed.
3. The bike was actually moving beforehand. The system looks back up to 10 seconds before the impact to confirm the bike was in motion. This catches cases where a stationary or parked bike gets knocked, dropped, or otherwise jolted— a genuine impact reading, but not a riding crash— and excludes it from the count.
Confirmed incidents are described using three independent dimensions: impact intensity (Minor, Hard, or Severe), accident type (Stationary Fall, Low-Speed Fall, High-Speed Fall, or Unclassified), and outcome (Resolved or Unresolved). Impact intensity is derived from peak_g, the peak absolute lateral acceleration during the impact, where 'minor' is < 8 g, 'hard' is 8–11 g, and 'severe' is > 11 g. Accident type is based on the bike's estimated wheel speed immediately before impact. Outcome describes whether the bike resumes after the stop; an unresolved incident is one where the wheel does not resume during the recorded route.
Filters can be activated individually or in combination. Each combination reflects a deliberate analytical perspective chosen by the user.
The dashboard does not prescribe a single interpretation, but supports exploratory, policy-driven analysis.
Behavioural indicators should be understood as signals rather than definitive diagnoses. Results should be interpreted alongside contextual knowledge of street design, traffic conditions, and policy objectives.