Sam - Outlier removal for visual odometry using Local Flow Consistency. Tim - Glasshouse Automation
Speaker
Sam Schofield & Tim Rensen
Institute
University of Canterbury
Time & Place
Thu, 02 Jun 2022 12:00:00 NZST in JE239
Abstract
Sam - Visual odometry is the process of estimating the motion of a camera using images that the camera is capturing. VO is often used in robotics to provide the robot with a pose estimate when GPS is unavailable or does not provide enough precision. Traditionally, VO relied on the assumption that it was operating in a static environment. Recently there has been an abundance of work aiming to improve the robustness of VO in dynamic urban scenes. However, unlike urban environments, dynamic vegetated environments remain relatively unstudied yet are essential for expanding robotics into areas such as agriculture and forestry. We propose a method for detecting (and removing) dynamic features by exploiting the differences in optical flow patterns caused by moving vegetation and camera motion. Our results show that the proposed method improves the accuracy and significantly reduces the processing time of visual odometry in dynamic vegetated environments.
Tim - 1/3 of my time is spent contracting to a horticulture automation company, FTEK, who manufacture personnel lifts and sprayers for capsicum, tomato and cucumber hothouses. We also recently showcased a more general crop maintanence robot.
Recording of both Sam and Tim: