Hi everyone! We’re developing a YOLO-based traffic monitoring system in Digos City to detect helmetless and triple-riding violations while preserving privacy (only logging time, location, and counts—no faces or plate numbers). We’re deciding between using a Raspberry Pi 5 for full on-device processing (detection + logging), which may face thermal throttling and FPS drops, or a client-server setup where cameras stream to a central server for processing, which may introduce latency and bandwidth issues. For real-world deployment, which approach is more reliable, and is the RPi 5 with NCNN sufficient for real-time detection, or should we consider accelerators like Jetson Orin Nano? Also, are there better optimization tools and best practices for strict privacy-by-design?