Hi NXP Support Team,
I am currently developing a real-time basketball pose and gesture recognition application on the i.MX93 platform running Yocto Linux (version 6.6.3). I have previously worked with various ML models on this platform, including driver monitoring systems (DMS), face recognition, and object detection, and have experience handling model integration and camera input pipelines.
For my current use case, I require camera-oriented gesture recognition specifically tailored to identifying and analyzing basketball actions—such as shooting, dribbling, passing, and defensive stances—in real time. The system should be able to continuously monitor players’ poses via the camera and provide live feedback based on recognized gestures or actions.
I already possess a relevant model file for this use case, but I would appreciate NXP’s guidance on the following:
Recommended ML framework or pipeline for integrating gesture-based action recognition using the NPU on i.MX93.
Best practices for preprocessing video frames for pose/gesture detection, especially under varying lighting and movement conditions typical in basketball environments.
Real-time feedback mechanism implementation strategies within the constraints of the i.MX93 hardware.
Example applications or demo references (e.g., from eIQ or TFLite models) that may help accelerate development for this use case.
Any additional BSP configuration or Yocto layer recommendations to optimize performance and camera integration for this task.
I look forward to your expert advice on enabling this solution efficiently on the i.MX93 platform.
