基于Harris-SIFT的移动机器人视觉定位  

Visual Localization for Mobile Robot with Harris-SIFT

作　　者：袁亮[1] 

机构地区：[1]新疆大学机械工程学院,乌鲁木齐830047

出　　处：《组合机床与自动化加工技术》2014年第6期19-24,共6页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：国家自然科学基金(61262059);新疆优秀青年科技创新人才培养项目;新疆大学博士启动基金

摘　　要：目标识别和深度估计是移动机器人视觉定位的两个难点问题。针对这两个问题,文章提出了Harris-SIFT特征提取算法和基于尺度空间的深度估计算法。Harris-SIFT结合了SIFT(Scale Invariant Feature Transform)算法和Harris角点检测器,去除SIFT得到的不具有显著角点特征的特征点,以提高SIFT特征点集合的整体显著性,从而改善匹配和识别效果。此外,Harris-SIFT只需要为部分SIFT特征点生成特征描述,缩短了计算时间,适合实时应用场合。基于尺度空间的深度估计算法通过计算参考图和目标图的特征尺度比,得到图像中同一目标的近似尺寸比例,再结合参考图中目标的深度信息,便可恢复出目标图中目标的深度信息。实验表明,在移动机器人自主导航过程中,基于Harris-SIFT的目标识别体系可以准确而有效地识别目标,同时尺度空间深度估计算法也能较准确地定位目标。结合Harris-SIFT和尺度空间深度估计算法可以很好地完成移动机器人视觉定位。Object recognition and depth estimaiton are the two key problems in visual localization for mobile robot. To solve these prolbems, this paper presents a feature extraction algorithm called as Hanis-SIFF and depth estimation algorithm using scale space. The Harris-SIFT algorithm is combined with the SIFT （Scale Invariant Feature Transform） algorithm and Harris corner dectector. It gets rid of some feature points with non-remarkable corner features in order to improve the whole significance in SIFT feature point collection and better the match and recognition performance. In addition, Harris-SIFT can be used in the real-time cases because it only take case of some of SIFF feature points and the computation time is decreased. The depth estimation algorithm based on scale space achieves the approximated dimensional scale by computing feature dimensional scale in a refernce image and objective image. Then the objective deption information in the objective images can be achieved by combining the objective depth information in the reference image. Experiments show that Harris-SIFT can accurately and quickly recognize the object in the navigation of the mobile robot. Meanwhile, the depth estimaiton algorithm in space scale also can localize the object accurately. Therefore, the visual localization for mobile robot can be improve by combining the Harris-SIFT and the depth estimaiton algorithm in the space scale.

关 键 词：Harris-SIFT 尺度空间 深度估计 视觉定位 

分 类 号：TH166[机械工程—机械制造及自动化] TG65[金属学及工艺—金属切削加工及机床]