Skip to main content

Hi Everyone,

I'm excited to be given the opportunity to work on ANPR-for-All, which aims to provide an easily-insertable automatic number-plate (vehicle license plate) recognition (i.e. ANPR also known as ALPR) engine for a variety of applications, such as for automatically opening gates for specific vehicles in emergencies.

This is my first time working with Axelera hardware and software, and I've gone ahead and installed the software development kit (SDK) onto a Linux box, and explored it a little. It was nice to see the install went smoothly.

I spent some time over the last couple of days taking around a hundred photographs of vehicles to label, so that I can tweak the dataset a little to the license plates in my region (UK). 

To speed up this task, I've written a simple image editor (as shown in the screenshot above) with keyboard shortcuts to quickly crop images, create bounding boxes and auto-generate label files. I'm still working on it, but will of course share it once it's in a usable state. Although it will be far less feature-rich compared to off-the-shelf tools, it should be faster to use since it will be easy to modify the code for specific tasks, and every second counts when working with large volumes.

Thanks for reading.

Hi ​@shabaz, great to see you here!

Oh wow, you’ve made some excellent progress already! Totally ninja! 😃 Love the idea of doing some prep work around your own data set training too. As you say, it’s exactly that sort of thing that can really help you gather momentum and speed things up when it really matters.

I hadn’t really considered how license plates look different in different regions, too. That’s a really interesting complexity that this kind of project is up against. Actually, I bet all projects have these kinds of complexities, but it’s awesome that you’ve identified it so early on in the build!

Gonna be amazing, dude! Rock on!


Reply