Vehicle speed estimation from video
Built independently with a senior colleague, then integrated at ODOT
A working roadside pipeline from video to metric speed — with accuracy never documented, and not estimated here.
Speed accuracy was never documented
The problem
A camera sees pixels moving. Turning that into a speed in miles per hour requires knowing how pixels map to the road surface, and doing it at the roadside on hardware that fits in a box.
The data
Real roadside video captured for the independent build.
At ODOT, the existing statewide camera network, which already produced bounding boxes and vehicle class labels.
Approach
Built first, independently: YOLOv8 detection with DeepSORT multi-object tracking and OpenCV frame processing, deriving speed from tracked movement over time, with real-time inference deployed to a Raspberry Pi camera unit at the roadside.
Then integrated at ODOT. Their pipeline already detected and classified vehicles; the contribution there was metric speed estimation via perspective transformation onto the road plane, plus tracking.
Results
A working pipeline from video through detection, tracking and movement measurement to speed estimation, with detections, track IDs and motion overlaid on the video.
Speed accuracy was never documented. There is no validated error figure for this system, and none is estimated here — a plausible number would not survive the first follow-up question.
Limitations and what I'd do differently
The missing accuracy figure is the limitation. Without a ground-truth reference — a calibrated radar or a known-speed pass — the system produces numbers nobody can check. Collecting even a small validation set would have been a day’s work and would have changed what this project can claim.
Perspective transformation assumes a flat road plane and a fixed camera. Grade, camera shake and lens distortion all enter directly as speed error, and none of them are corrected.
Tracking identity switches under occlusion produce speed outliers. There is no filter on implausible accelerations.
Artifacts
Part of 480+ documented research hours at ODOT across three assignments, with three milestone presentations and a final technical report.