So this is the final day. There was not much I did today. I only cleaned up my code and did some minor fixes/improvements. I am happy that the first part of my project kind of succeeded. Of course there are some problems left (e.g. grey/white traffic signs are only found in around 50%), but basically the software works at min. 1 FPS on CPU and is therefore usable under real conditions. I also created a project page for the whole BADASS Chauffeur Software. On this page I will also add my current sources, the trained CNN and 2-3 more result images.
I am going to add a short video to this project page as soon as I am having more time. I am looking forward to the extension of my sign-detector and to the other parts left, which I also listed at day #1. Hopefully you follow the project in future! 🙂
The first part was more challenging than I thought it would be. Especially the very limited hardware, which I wanted to use, made the problem quite hard to solve with state-of-the-art solutions (like Faster RCNN). Even though there were days, where I really had no idea what to try next, it still was a lot of fun to try different (and sometimes really strange) ideas. I learned much in this first part. Not only regarding pattern recognition and CNN but also regarding webcams and smartphone-cameras. So overall it was a really nice experience! I also want to say thank you to everyone who helped me so far, especially when annotating images. When I release the dataset, I will also add a list of every person who helped me creating/annotating these images.