Harnessing the power of Raspberry Pi 4 to build cutting-edge computer vision solutions. Whether you're interested in object detection, image classification, or real-time video analysis, this project will give you the tools you need to get started.
MIT License
0
stars
0
forks
source link
Evaluate Traffic Counter using ByteTrack and ncnn detector on test videos #21
Better to test on RPI directly! Why?
a) Extensive setup required, NCNN is setup but eigen + thread is not installed and documentation limited
b) Uses libcamera which is a rpi4 mdule (Extensive code rewrite required otherwise)
c) C++ project - Python preferred by new devs.
Things to take forward:
1) Nanodet a possible solution -
Good points: Fast and accurate
Bad points in Pytorch , we want to run on rpi so either conversion or some other caveats
2) Bytetrack have been used and seem to do the job
Good points : Fast and most accurate
Bad points : Unknown territory, Study required + Conversion from C++ to py code from here is a starting point.
Suggestions:
1) YoloV4-tiny-tflite
Good Points: Fast , accurate, optimized for embedded devices, catchy
Bad points: Model not found yet + Speed questionable + tracking dubious in skipping framerate
Better to test on RPI directly! Why?
a) Extensive setup required, NCNN is setup but eigen + thread is not installed and documentation limited b) Uses libcamera which is a rpi4 mdule (Extensive code rewrite required otherwise) c) C++ project - Python preferred by new devs.
Things to take forward: 1) Nanodet a possible solution -
Suggestions: 1) YoloV4-tiny-tflite