SparrowMap: Cameras That Watch Government Cars

The SparrowMap project offers an unusual answer to the familiar logic of video surveillance: instead of cameras watching citizens, these cameras watch government cars. The slogan sounds fitting — “watching the watchers.”

The scheme is voluntary and utterly simple. A participant points an unneeded, already retired phone out a window facing the street. License plate and vehicle type recognition happens right on the device, in an open browser tab — with no app to install and no registration. It works on an old smartphone, on a laptop's webcam, and on a USB camera.

What Is Kept and What Is Erased

The key idea lies in the asymmetry of data. If the camera sees a government car, its photo and plate number are saved and become part of the public record. The authors frame this as a principle: public money, public roads, public transport.

Everything else is destroyed. An ordinary car's plate number is erased right on the camera, before anything ever leaves the device. Only an anonymous point is sent to the server — with no photo and no plate number. This approach also removes the main surveillance risk: since the video stream is never transmitted anywhere, there is nothing to intercept. Only the recognition results are sent to the server, not the video itself.

The authors also emphasize other measures to protect the participants' own privacy: a camera's exact location is never published — only the street it is watching is given. You can turn the camera off and take it offline at any moment.

A Permanent Camera

For those who need an “always on” mode, there is an option using a spare computer and a webcam: a full-fledged detector is deployed on it, one that locally distinguishes a patrol car from an ordinary one and reads plate numbers. Installation comes down to a single command — for Windows it's a PowerShell script or a standalone installer, and for Linux a curl-based command is offered, though the developers are still asking for help testing Linux support.

In every case, the same principle recurs: the computation stays on the user's machine, and only the recognized objects go out, not the video recording. The network has no owner, the authors claim — it rests on volunteers, each of whom connects and disconnects their own camera on their own.

Source: Hacker News