Can object detection and pose recognition be done simultaneously on imx8mplus??

[dpcivl]

Hello, I'm a user working with an imx8mplus development board. I have developed programs for object detection and pose recognition using Python. For object detection, I used the ssd-mobilenet tflite model, and for pose recognition, I used the movenet tflite model.

However, I am facing a problem where the memory allocation fails because the code defining the capture for object detection and pose recognition overlaps.

```cap = cv2.VideoCapture('v4l2src device=/dev/video2 ! video/x-raw, width=640, height=480, framerate=30/1 ! videoconvert ! appsink', cv2.CAP_GSTREAMER)```

I am using the above code in both the object detection and pose recognition programs, but it seems to cause issues since one camera is being used for two GStreamer pipelines simultaneously. Is there a way to perform detection for both object recognition and pose estimation at the same time using a single camera?

[adarshkv]

Hi! Thanks for sharing the details of your setup. It’s really interesting that you’re running both object detection and pose recognition on the i.MX8M Plus board.

I also tried using the MoveNet TFLite model for pose detection, but I wasn’t getting proper results — maybe due to some issues in inference or model behavior on the board imx8mp.

If you don’t mind, could you please share the Python code or a sample snippet that you used for MoveNet pose detection? It would be helpful to compare and understand how you’ve implemented it.

Thanks a lot!

[brian14]

Hi @dpcivl, 

Thank you for contacting NXP Support.

Unfortunately, I couldn't find any example or demo that is tested and guaranteed on our side.
However, I found that you can apply an asynchronous execution in your python script to apply one model and after this model have finished you can execute the second model.

Please have a look on the following reference:
concurrent.futures — Launching parallel tasks — Python 3.12.1 documentation

Have a great day!