SIMPLE AUTOMATIC PATH LEARNING FOR AUTONOMOUS VEHICLE NAVIGATION BY ULTRASONIC SENSING AND COMPUTER VISION TECHNIQUES

Abstract

A simple automatic path learning method for an intelligent autonomous vision-based vehicle system by person following and along-path object image matching is proposed. The vehicle can follow a person through a path in an indoor environment and navigates back to the start point by itself by the data learned during the following process. A technique of matching flat-surfaced object images using SIFT features is adopted for the vehicle to localize itself according to its learned path during the navigation process. In addition, a technique for dynamically refining the traversed path points which are learned during person following is proposed. Finally, a technique of odometer calibration, which uses a calibration model to reduce incremental mechanical errors the vehicle suffers, is proposed to increase path traversing accuracy. Good experimental results show the feasibility of the proposed techniques.