I want to estimate 3D pose of camera between consecutive frames in unknown environment, this should run on handheld camera like Android or iPhone.
I have camera intrinsic and some feature points detected in frames
and there is no extrinsic parameters of camera or predefined real object in environment for tracking.
At first stage performance does not matter
Currently I'm looking for a simple way like following steps:
- Get new frame.
- Detect feature points in new frame.
- Detect match points in new frame and old frame.
- Extract translation and rotation from match points.
Now my question is:
1-How I can detect 3d pose changes of frames in simple way with my requirements?
2-If it's not possible, should I implement this in complex methods like
Trajectory,
SLAM,
SFM,
Visual Odometry or Egomotion.
Or any other better solution?