[robotics-worldwide] [software] Release of SVO 2.0 (Semi-Direct Visual Odometry) - Runs up to 400fps (100fps on smartphones) on less than half a CPU core
we are happy to release the binaries of SVO 2.0, the latest update of
our Semi-direct Visual Odometry pipeline (originally open sourced in 2014).
As its predecessor, SVO 2.0 runs up to 400 fps on an Intel i7 or up to
100 on a smartphone processor (e.g., Odroid) while taking less than half
a CPU core! Compared to the original open source implementation
(ICRA'14), SVO 2.0 (TRO'17) adds IMU priors, edgelets, extension to wide
angle cameras (fisheye and catadioptric) as well to stereo camera
systems. Comparison of SVO 2.0 with respect to other state of the art VO
algorithms or semi-dense and dense approaches can be found in the
reference paper (TRO17).
SVO has been used in our lab for the past 4 years as the main
vision-based state-estimation engine of all our vision-based autonomous
drone projects, but it has also been used in other projects, including
automotive and mobile devices (AR/VR).