NanoDet on RKNN2

Python Version

Based on rknn-toolkit2 and rknn-toolkit-lite2
use OpenCV for image capture & process
- Capture image // TODO: Set attributes of capture stream
- Resize it to (320, 320) and convert to RGB
- Feed converted image to RKNN, get result of inference
- Parse result, draw box as overlay of origin image // TODO
- Async / Multithread support maybe? //TODO

CPP Version `// TODO`

Based on rknn-rt2
Use V4L2 / GStreamer / librga / mpp...?
- Maybe capture YUYV, then convert YUYV to NV12?
Other, TBD...