qengineering / hand-pose-ncnn-raspberry-pi-4 Goto Github PK

I followed the installation steps and got the code to work.
But without changing a single line and therefore still using the example hand.jpg image,
the predictions don't seem to fit:

Is this a known issue? Doesn't look like a scaling issue

Hi ,
@Qengineering

Thanks for your quick reply and example for help . Really.

I tested below image and video nothing detected.

vide green square was at original

Best
👍

PS: Debugging gives error on Raspberry Pi
have experienced on the Code::Blocks

Hi

HW:

• Raspi4 4GB
• Raspi Camera Module V2 8MP

I installed the App from this Github repo and it shows the picture with the detected Hand

now when I run a Camera stream (with Gstreamer) it displays the Camera stream a little laggy (definately not 30fps, rather around 5fps), but it works.

I got 3 questions now

1. (not very important)
   How do I improve the raspicam's fps? (I've seen it way smoother than using Gstreamer using this WebInterface here)

2. (the more important question)
   How do I merge the App recognizing my Hand's Finger count / gestures with the live stream of the Camera, instead of just the picture?
   I saw the example where it says "YoloX + GStreamer camera." but I don't quite understand it, maybe someone could help me? :)

3. (also important)
   How would I get the count of the fingers shown in the live video feed?
   Like is there some variable it's stored in, or do I have to calculate it myself?
   I totally not understand how I can use these Data from the App for other things (f.e to change the Colors of a RGB LED Stripe according to the number of fungers shown.

.

Thank you verry much for your help in advance :)
Regards, Lixx

如图所示，本人在一块ARM64的板子上部署了你的代码，但是它并没有正确识别到，请问这种情况需要我怎么修复？