NECEC 2017 has ended
View analytic
Wednesday, November 15 • 15:40 - 16:00
Computationally-Efficient Visual Inertial Odometry for Autonomous Vehicles

Sign up or log in to save this to your schedule and see who's attending!

The objective of this research is to develop a computationally-efficient monocular Visual Inertial Navigation System applied for autonomous vehicles and wearable devices, which are subject to limited computational resources and electrical payloads. The traditional system executes a complicated recursive 3D feature-point reconstruction step to correct the prediction from IMU data and produce the reliable estimate of the vehicle's ego-motion. This, in turn, slows down the estimation process and decreases the system stability. Those issues will be addressed by using Trifocal Tensor Geometry to replace the expensive image-processing algorithm. An economic sensor fusion algorithm is proposed to decrease significantly the computational cost, and improve the system consistency. It is a combination between cubature Kalman filter, multi-state constraint Kalman filter, and information filter. Experimental validation is conducted to evaluate the proposed system.

Wednesday November 15, 2017 15:40 - 16:00
Salon A/B 180 Portugal Cove Rd, St. John's, NL A1B 2N4, Canada

Attendees (1)