基于视觉特征的移动机器人Monte Carlo定位方法

提出了一种通过引入基于三角定位的重采样阶段对基于视觉特征的常规Monte Carlo定位加以改进的方法，以提高原有方法的实现效率，既能提高计算效率，又能避免过收敛现象。重采样的实现根据感知更新前后采样分布信息熵的变化和有效采样数目来判断，并且基于感知组织的贝叶斯网络识别视觉特征的方法为三角定位提供了准确的特征来源，有效减少了假阳性特征，大大简化了与环境模型的匹配。实验结果验证了方法的有效性。

Vision-based Monte Carlo localization for Mobile robot

Standard vision-based Monte Carlo localization for mobile robot is augmented with triangu- lation-based resampling process introduced to increase computational efficiency and avoid over-convergence in this paper. Bayesian networks based perceptual organization is de- veloped to detect features in office environments, which provides reliable features for trian- gulation-based resampling process, avoids false positives and simplifies correspondence between sensing and models. Experimental results demonstrate the validity of the approach.