基于全景视觉匹配的移动机器人蒙特卡罗定位算法

摘要

We propose an algorithm to solve the omnidirectional image matching problem by using the histogram of gradient orientation. By combining the matching algorithm with Monte Carlo localization algorithm, we develop a method of localization for the mobile robot based on the omnidirectional vision. The characteristics of the matching algorithm are analyzed, from which we build the observation model and develop the method for calculating the important weights for the particle filter. This method rejects the interference of perceptual aliasing during the localizing period, and helps the robot to rapidly recover from the “kidnapped”situation. Experimental results show the validity and effectiveness of the presented method.%本文提出了一种基于梯度直方图的全景图像匹配算法，并将该算法与蒙特卡罗定位方法相结合，构建了一种基于全景视觉的移动机器人定位方法．在分析所提出的匹配算法特点的基础上建立了系统的观测模型，推导出粒子滤波中重要权重系数的计算方法．该方法能够抵抗环境中相似场景对于定位结果的干扰，同时能够使机器人从“绑架”中快速恢复．实验结果证明该方法正确、有效．