I know, there are already tons of "Tello face tracking" and even some pose tracking videos out there. I decided to go for my own aproach:
Compared to face detection, pose detection allows a better interaction with the drone and also works with backlight.
Implementation is based on DJITelloPy and OpenCV (Python version) with TensorFlow. It was quite difficult to find a pose tracking model, that also runs with satisfying frame rate even on slower hardware without GPU acceleration. I decided for a pretty simple approach. There are more accurate pose detection models out there but most of them are too slow without powerful GPU hardware.
At the moment, it only follows the nose using by turning on the yaw axis. For a smoother "nose tracking", PID control is applied. Next step will be to make it also horizontally follow the detected person...
Compared to face detection, pose detection allows a better interaction with the drone and also works with backlight.
Implementation is based on DJITelloPy and OpenCV (Python version) with TensorFlow. It was quite difficult to find a pose tracking model, that also runs with satisfying frame rate even on slower hardware without GPU acceleration. I decided for a pretty simple approach. There are more accurate pose detection models out there but most of them are too slow without powerful GPU hardware.
At the moment, it only follows the nose using by turning on the yaw axis. For a smoother "nose tracking", PID control is applied. Next step will be to make it also horizontally follow the detected person...
Last edited: