0

I am working on a system that relies on MediaPipe's pose detection to track a person's movement. It currently continuously processes frames from a video stream.

The issue is the landmark predictions sometimes move to another area of the screen suddenly. I am hoping to have landmark predictions that are conditional on previous estimates to avoid this problem.

Does MediaPipe have a way to support this? If not, how could I implement it?

Relevant code:

# Initialization
self.mp_poseObject = mp.solutions.pose
self.pose = self.mp_poseObject.Pose()
...    
# For each given frame from the video feed:
poseResults = self.pose.process(self.videoFrame_RGB)
self.landmarks = poseResults.pose_landmarks
Christoph Rackwitz
  • 11,317
  • 4
  • 27
  • 36
dylan
  • 43
  • 1
  • 5
  • i hope those tags help summon the right demons – Christoph Rackwitz Apr 22 '23 at 19:43
  • 1
    Have you tried playing around with the parameters `min_detection_confidence=0.5` and `min_tracking_confidence=0.5`, `model_complexity=2,` (https://github.com/google/mediapipe/blob/master/mediapipe/python/solutions/pose.py) .. You can try different combinations to see if that improves anything., – Atif Anwer Apr 26 '23 at 12:19

0 Answers0