RA-L2021: Real-time rotatinoal motion estimation with contrast maximization over globally aligned events

- 2 mins

SNU LARR

Abstract

Contrast maximization is an event camera application that can estimate angular velocity, depth, and optical-flow using a subset of events observed in a temporal window. In the estimation of rotational motion, we can compute the angular position by integrating the angular velocity. However, the accumulation of drift error degrades the accuracy of motion estimation. If the contrast maximization framework utilizes events measured before the temporal window, the performance of the framework will be improved, including the alleviation of drift error in motion estimation. In this work, we utilize the globally aligned event data and propose the rotational position and velocity estimation method using an event camera only. The proposed algorithm not only maximizes contrast of an image of events in a single temporal window but also maximizes the contrast image of events observed over time. Our algorithm works in real-time by reducing additional computations of the existing contrast maximization. We confirm the real-time operation with a single-core CPU on a laptop and show that the maximum error is within 3 degrees on public data sets and acquired real-world data sets. To contribute to the community, we provide the source code and the real-world data sets to the public.

Video

Source code

https://github.com/Haram-kim/Globally_Aligned_Events

Datasets

Data sets with VICON ground truth pose

sequence name description cx cy fx fy dist_coeff1 dist_coeff2

360_indoor
ZIP(29.4MB)
360_indoor 230.2097 231.1228 121.6862 86.8208 -0.4136 0.2042

fast_motion
ZIP(23.7MB)
fast_motion 230.2097 231.1228 121.6862 86.8208 -0.4136 0.2042

ESIM_panorama
ZIP(30.5MB)
ESIM_panorama 200.0000 200.00 120.0000 90.0000 0.0000 0.0000

ESIM_OpenGL
ZIP(36.3MB)
ESIM_OpenGL 200.0000 200.00 120.0000 90.0000 0.0000 0.0000

Data sets without ground truth pose

sequence name description cx cy fx fy dist_coeff1 dist_coeff2

360_outdoor
ZIP(74.9MB)
360_indoor 230.2097 231.1228 121.6862 86.8208 -0.4136 0.2042

fast_roll
ZIP(25.1MB)
fast_roll 230.2097 231.1228 121.6862 86.8208 -0.4136 0.2042

lobby
ZIP(35.7MB)
lobby 230.2097 231.1228 121.6862 86.8208 -0.4136 0.2042

rooftop
ZIP(49.2MB)
rooftop 230.2097 231.1228 121.6862 86.8208 -0.4136 0.2042
rss facebook twitter github gitlab youtube mail spotify lastfm instagram linkedin google google-plus pinterest medium vimeo stackoverflow reddit quora quora