Semi-dense Visual Odometry for a Monocular Camera

Semi-Dense Mapping

Stereo to find depth information

The idea is to use small baseline stereo where we can

Depth map update algorithm

  1. Find pixels with high disparity accuracy
    • They identify 3 criteria for this
  2. For each pixel, select a suitable reference frame, and do a 1-dim disparity search on the epiline
  3. Use priors where possible to make this fast (reduce the disarity search space)
  4. Fuse observations into the current depth map

Reference Frame Selection

Stereo matching

Uncertainty Estimation

Depth map fusion

Summary

Depth map propagation

\[d_1(d_0) = (d^{-1} - t_z)^{-1}\]

Depth map regularization

Local smoothing

Outlier removal

Dense Tracking

\[E(e) = \sum_i \frac{\alpha(r_i(e))}{\sigma_{d_i}^2}r_i\]

System Overview