If you are not sleeping with a blanket, I'm suggesting either video recording yourself (e.g. with a cellphone) and after that running a pose estimation model over the video, or running the model live (this one seems promising: https://www.tensorflow.org/lite/examples/pose_estimation/ove... ), if you are sleeping with a blanket, the model may need training. Then once you have the pose estimations, it should be easier to analyze. Please contact me if you want us to try together.