关键词: 定位, 粒子滤波, 相似度, 核函数

Abstract: To improve the localization capability of the mobile robot in a dynamic and unstructured environment，a mono-vision based particle filter localization approach is proposed. The localization is achieved by the computation of the similarity between query images and the images stored in an image database.After the extracting of the vision features which have the properties of invariance of translation，rotation and scale，a relational kernel function is applied to reduce the effect cased by environment noise and variations in illuminations.Based on the local features，the similarity is used as the weight of the particle and the post probability of the particle is updated by the visible areas of the reference images.The practical experiments illustrate that the approach is able to locate the robot accurately and effectively under the dynamic environment in which the geometry features are extracted difficultly.

Key words: localization, particle filter, similarity, kernel function