Motion Estimation using Optical Flow

1
Motion Estimation using Optical Flow

• Tarak Gandhi
• Mohan Trivedi

2
Motion Estimation

• Apparent motion observed in image sequence is
  known as optical flow.
• Under favorable conditions (brightness
  constancy), optical flow constraint is satisfied
• Space gradients gx ,gy
• Time gradient gt
• Optical flow ux, uy
• One equation, two variables Aperture problem!

3
Aperture Problem

• No motion information in uniform regions.
• Only normal component of motion available on edge
  points.
• Corner points give full motion

4
Motion Estimation Methods
5
Corner Based Estimation Lucas-Kanade

• Write optical flow equations for each point i,
  assuming constant motion (ux,uy) in neighborhood.
• Multiple equations in two variables. Solve
  equations using least squares.
• Equations well conditioned only near corners.

6
Hierarchical Motion Estimation

• Estimation based on optical flow estimation fails
  for large motions
• Use iterative coarse to fine (pyramid) method
• Motion parameter estimates at coarse resolution
  used to warp one image towards another at finer
  level
• Residual motion computed using finer level images

7
Hierarchical Motion Estimation
8
Kanade-Lucas-Tomasi (KLT) Feature Tracker

• Public domain software for optical flow based
  feature tracking.
• http//www.ces.clemson.edu/stb/klt/index.html
• Extracts corner-like features.
• Finds optical flow at these features using coarse
  to fine estimation
• Verifies reliability, track over frames
• Replaces lost features at each frame.

9
Results
Actual feature motion (red) Residual feature
motion (yellow)
Estimated feature motion
Only forward moving persistent features
classified as obstacles
Classification Red forward, Green
stationary/slow, Blue backward
10
Gradient Based Estimation

• Motion of road plane given by planar motion
  model
• Substitute this in the optical flow equation
• Solve using least squares or Bayesian estimation.

11
Block Diagram
12
Experimental Results
Estimated parametric motion
Features Black unused, Gray pure inliers,
White pure outliers
Normalized difference image
Tracked objects with track ids and positions
13
Conclusion

• Image motion satisfies optical flow constraint
  under favorable conditions
• Due to aperture problem full image motion cannot
  be estimated at all points
• Corner based motion detection
• Computes image motion at corner-like features
• Gradient based motion estimation
• Uses parametric motion model over larger parts of
  image and directly estimates the parameters using
  optical flow constraint