Abstract:

In order to improve the measurement accuracy of a monocular vision-based bidirectional automatic guided vehicle,a motion-based self-calibration method is used to reduce the system error,and a color image processing is performed to extract the centerline of the guided path. Then,a classification method of path models based on curvature estimation is proposed to meet the requirements of target accuracy. The path models are classified into three types,namely the straight line,the arc turning and the non-circular turning. Finally,the minimum mean-variance method is used for the parameter regression of the first two models,and the curvature estimation-based adaptive weighted fitting method is used for the parameter regression of the third model. Experimental results show that the proposed method is of high measurement accuracy and meets the requirement of vision navigation well.

Key words: vision navigation, automatic guided vehicle, curvature estimation, adaptive weighted fitting