Camera board • car-speed measurement project

In case of interest, here's a car speed measurement project using a Pi Global Shutter camera looking perpendicular to the road. Speed is estimated both from an overall sliding template match, and the median of a set of individual keypoint matches (filtered for a consistent motion vector), between two frames taken 50 msec apart. Knowing the image pixel scale requires accurate camera-vehicle distance, which here I acquire separately with a lidar. It could be estimated from the image directly, if you can locate the point in the image where the tires meet the road, and then apply a pre-calibrated lookup table. That assumes a fixed camera location and FOV.

This sort of thing is nothing new and I have seen similar projects years ago, but I had fun building it. The more common cameras have a pretty significant rolling-shutter artifact on faster cars, that I found annoying visually. But in this application it would probably not affect the measurement, as each frame would have the same distortions.

Github project: https://github.com/jbeale1/traffic

Nearly 200 keypoints might seem excessive, but more data = better averaging, and compute power is cheap these days.

Statistics: Posted by jbeale — Thu Jun 18, 2026 6:56 am

---