Imagine you have a sensor module on your desk. It is plain but affordable. In your drawer you have several MV cameras; fully featured, not so affordable. But they both use the same image sensor.
In the MV camera you pay for the sensor integration, the calibration, the features, and the reliability. What if you could have all of that from the sensor module as well?
This is what I'm aiming for with libvidere! If you want to help bring this vision to life, you are very welcome!
Wouldn't it be great if the plain module behaved like the expensive camera?
MV cameras don't justify their price with a better image sensor. They justify it with everything wrapped around the sensor: the integration that turns the sensor into a clean stream, calibration you can trust, a predictable set of features, and the reliability to lean on in production.
A bare CSI-2 module gives you none of that. It hands you pixels and then leaves you alone. The sensor is capable; the software around it is usually as bare as the module.
Wouldn't it be great if you could just ask for a feature and get it?
What do you actually want from a real camera? A hardware trigger and a strobe output? Maybe per-frame metadata that tells you the exposure and gain that were actually applied? Color you can trust? And maybe several cameras that stay in sync?
On commodity hardware, each of those can be its own little research project. Some sensors have a trigger line. Some give you usable metadata. Usually, each sensor does it differently.
With libvidere we aim to enable all of those features, whether the sensor offers them or not.
Wouldn't it be great if the API never made you care how?
In libvidere, the same call behaves the same way whether the feature is native to the sensor, emulated in software, or synthesized by the image pipeline. A small profile describes what each sensor can do on its own, and the runtime quietly routes your request to wherever it can actually be served.
Ask for a trigger. If the sensor has the line, it goes to hardware. If it doesn't, it becomes a software trigger. You can write it once, and it just happens. You never branch on the sensor. You never rewrite your code when you change modules. Timings may change but we aim to make that transparent as well.
That's the whole trick: sensor differences stop being your problem and become libvidere's.
Wouldn't it be great if machine-vision behavior were just software you could install?
That's the bet.
We are not doing a new sensor or a new board. We are just building a runtime that takes the capable, affordable hardware you already have and gives it the integration, the calibration, the features, and the reliability you'd otherwise have to buy.
That's what we are aiming for with libvidere. If that gets you excited as well and you'd like to help bring it to life, please reach out and you're very welcome.