VISION SENSOR POSITION FOR GUIDANCE BY USING POSE-GUIDANCE ALGORITHM (PGA)

摘要

A key concern when developing an autonomous vehicle based on a vision sensor are sensor placement and sensor field-of-view. This research addressed these issues using a pose-guidance algorithm (PGA). PGA encoded images using principal component analysis and classified vehicle pose (offset and heading angle) by running the minimum Euclidean distance classifier. PGA performance was studied using 9 combinations of three different sensor heights and tilt angles, as well as 5 different image regions of interest. The best PGA performance was reached with sensor height of 1.22m and tilt angle of 15 degrees. PGA reached the expected average of the absolute heading angle error of 0.5 degrees but did not reach the expected average of the absolute offset error of 0.2 cm. The average of the absolute offset error was 4.1 cm.