mlda-ntu / drone-vision Goto Github PK

Refer to this https://github.com/atomicbits/posenet-python
Generate converted posenet model for resnet (SavedModel format)
Use it instead of tf-lite

code should be refactored so that it's easier for drone-control team to use, instead of copying the code over to drone-control code

video stream from drone is not steady, and frame rate is not fixed. Thus, pose estimation should only be called when the next frame is received, instead of calling every time after the loop

from pose information, deduce static gesture (e.g. raise hand straight up, 90 degree...)
potential approach: calculate the angle between some lines connecting joints

obtain multi-pose from posenet output