ORIGINAL VIDEOPlayback · no server upload
Your local footage appears here
No source00:00
COMPUTER VISION · AI TRAFFIC MONITORING
Traffic insights,
powered by AI.
Watch vehicles get detected, tracked and counted, right on your device. No video uploads to our servers.
YOLO26ONNX · WebGPUPrivate by design
● ON-DEVICE AI · TRAFFIC INTELLIGENCE
Live traffic detection & analytics · powered by real YOLO inference
01Video frames
02AI detections
03Track & count
YOUR AI WORKSPACE
Start detecting.
Use a local video, live camera or a public direct video file. AI inference runs in your browser.
● Model not loaded
Use your webcam or connected camera in this secure browser session.
Only direct HTTPS MP4/WebM file links that allow CORS can be analyzed. YouTube videos cannot be analyzed from a YouTube watch link. Use Video upload with a local video you own or have permission to use.
YouTube video analysis isn't supportedUse the Video upload tab to select a local MP4/WebM video. Only download or use footage you have permission to use.
WAITING FOR SOURCE0%
PROCESSED FRAMES 0VIDEO POSITION 00:00VEHICLES VISIBLE 0
LIVE AI PREVIEWWaiting
AI bounding boxes appear here after inference
Selected vehicle
YOLO is not always correct. Choose the actual vehicle type and apply it to this tracked vehicle.
No AI frames yetCPU / WebGPU
Accuracy tip: Start with 0.5× playback to give the browser more time to process video frames. Use a 10 or 25 steps/sec Sequential scan for a more detailed (but slower) pass. These settings cannot create a new Van class or guarantee perfect detection. Review uncertain Bus/Truck labels by clicking a detection or crossing and choose Van (manual) if appropriate. Browser timeline seeking may skip/duplicate decoded frames.
Recent line crossings
Nothing has crossed the counting line yet.
UNDER THE HOOD
From a frame to an insight.
Real pretrained computer vision, with tracking and counting logic. These are the actual steps—not simulated detections.
01
Read a real frame
The video element provides current pixels from your chosen file or camera.
02
Run pretrained YOLO
An ONNX model predicts boxes, classes and confidence for road vehicles.
03
Associate tracks
Nearby matching boxes keep the same temporary vehicle ID across sampled frames.
04
Detect line crossings
When a tracked center moves across the horizontal line, the count increases. We stabilize labels using multiple observations, not one guess.
05
Inspect the results
See annotations, vehicle breakdown, crossing direction and exportable counts.
Want to learn the machine learning pipeline?
The developer Jupyter notebook includes class-confusion analysis and runnable Python examples: loading a real video frame, exporting YOLO to ONNX, tracking, counting and validation. The public website explains the concepts without publishing copyable code snippets.
ABOUT THE DEVELOPER
Built at the intersection of
AI and full-stack development.
VehicleAI is a computer vision portfolio project developed by Kanil Kanchana. Its goal is to make vehicle detection, tracking and traffic measurement accessible through a clean web interface.
VEHICLE DETECTION · TRACKING · INSIGHTS
BULLET & CONNECT
Help improve
VehicleAI.
Have an idea for a new AI feature, better counting, or a clearer dashboard? Send feedback directly to the project.