Eclipse Lamp Devlog #4 - Finally getting the Person Detection AI to run properly on my Raspberry Pi Camera

Today went EXTREMELY WELL in terms of progress. Finally, after what started back in Devlog 1 and Devlog 3, everything has come together and actually worked end-to-end in a way that feels complete. What began as a messy setup with camera errors, broken environments, and YOLO not even running correctly slowly turned into a full Raspberry Pi computer vision system with a working Picamera2 pipeline, a Flask web server streaming live footage to a browser, and YOLO detection successfully running on real-time frames. Along the way, there were a lot of setbacks—camera being “busy,” venv issues breaking imports, OpenCV display problems, dependency conflicts, and even moments where the system felt completely stuck—but step by step those were fixed until the full pipeline finally ran without crashing. By the end, I actually got YOLO detection working properly on live camera input, with bounding boxes showing up and the whole system functioning as intended. The main issue left now is performance: the system works, but it’s still slower than I want, with noticeable lag and lower frame rate than ideal. I went back and rebuilt parts of it trying to make it faster, but I didn’t fully finish optimizing it yet. That final tuning—improving FPS, reducing delay, and making the stream smoother—is something I’ll come back to later. Still, today was a complete success because the full system is alive, functional, and running exactly as planned, and now it’s just a matter of refinement rather than building from scratch.