Vision-aided inertial navigation

Abstract

A technique of removing inertial-odometer drift errors by identifying landmarks along the vehicle's route is presented. Landmark identification is performed with the use of laser range data. The scenario being developed consists of a ground vehicle equipped for autonomous retrace of a path that was stored while the vehicle was being remotely piloted to its destination. On the outbound portion of the vehicle's journey, objects are detected with the laser range imagery, tracked from frame to frame, and recorded as landmarks if they meet the requirements of the landmark model. During the retrace phase, inertial-odometer data are used to guide the vehicle along the outbound path (in reverse). The landmarks recorded on the outbound path are again located during retrace, and their locations serve to remove any positional error that has occurred. This technique of vision-aided inertial navigation is presented along with processed laser range data.< >