论文部分内容阅读
提出了一种基于视觉特征的移动机器人大范围地形感知方法.该方法采用由近及远的学习策略:首先,将获取的图像分割并进行尺度归一化,提取样本中反映颜色的色相特征和反映纹理的局部二值模式(LBP)特征作为描述子;其次,利用双目视觉将近景的一部分地形样本分为障碍与地面,将这些样本作为有标签的训练数据构建分类器分类未知样本;最后,基于后验概率定义可信度,对可疑的样本进行重分类,提高最终分类准确率.实验结果表明,该方法可以准确且稳定地实现大范围地形感知.
This paper proposes a near-distance and far-end learning strategy based on visual features of mobile robot. Firstly, the acquired image is segmented and normalized to the scale, and the hue characteristics of the reflected color in the sample are extracted. Secondly, using the binocular vision, part of the terrain samples of close-range landscapes are divided into obstacles and terrain, and these samples are used as tagged training data to construct classifier classification unknown samples. Finally, , Based on the posterior probability to define the credibility, reclassify the suspicious samples and improve the accuracy of the final classification.The experimental results show that this method can accurately and stably achieve a wide range of terrain perception.