Car
Private cars and similar light four-wheelers in the frame.
Compiling the hour
Compiling the hour
Vehicle detection
Upload a street or junction photo. YOLOv8 counts vehicles so density can sit beside history — useful when a camera sees what last Thursday’s average cannot.
YOLOv8
Camera location
Not selected yet
Type a place name. Results come from OpenStreetMap (Nominatim + Overpass), debounced while you type.
How to use this
Choose a clear photo of a Thiruvananthapuram road, flyover approach, or parking queue. Submit. Flask runs YOLOv8 and returns an annotated image plus class counts (car, bus, truck, motorcycle, bicycle).
A forecast knows the usual Thursday. It does not know a stalled bus ate two lanes at a gate unless someone tells it. A camera is that someone. Detection turns boxes into a density number the loop can use.
Detection on your uploaded image is live model inference for that file. City-wide map speeds elsewhere on the site remain demo forecasts from time and scenario. Do not mix “I counted 42 cars in this photo” with “every road in TVM is live right now.”
Counts are meant to align with the same idea as 5-minute traffic bins: how busy is this view. If YOLO is offline on a slim host, the API should say so; maps and forecasts can still run without it.
Not number-plate policing. Not a fine. Not a claim that every pole in Kerala is wired. It is a density instrument with a published architecture and a box you can argue with.
Prefer daytime, forward-facing junction views. Avoid heavy blur. Night and glare reduce recall. One frame is a sample, not a full hour of flow — treat it as a check on density, then return to Predict or Map for the city graph.
What gets counted
Private cars and similar light four-wheelers in the frame.
Two-wheelers — critical for Kerala streets.
Transit and large passenger vehicles.
Goods vehicles that change junction delay a lot.
Included when visible; not every frame will have one.
People, umbrellas, billboards — not treated as traffic density.
Common questions
Weights may be missing on a slim host. Forecasts and maps still work. Check /api/health for yolo: true/false.
This page accepts uploads. A production CCTV pipe would need privacy policy, retention rules, and a stable mount — not only a model.
Not automatically in the current demo loop. Counts are the density instrument; Map still uses scenario speeds unless you wire them together.
A daytime wide shot of a junction approach with visible vehicles. Cropped logos and night selfies under-count.