一种基于机器学习算法的单目里程计研究

摘要

Visual odometry (VO) through the trajectory calculation,cumulative motion vector,the relative positioning method of current position. Monocular odometer using only a single camera as image acquisition carrier, make information requirements lower,and can identify and locate the feature point accurately,real- time,cost much less, so it has more wide prospect of application. This paper uses SURF algorithm to simultaneously detect and match feature points,using a machine learning algorithm based on adaptive Calman filter (SVM),slow down will appear originally Calman filter in low precision and divergence,to system optimization of monocular odometric accuracy.%视觉里程计（VO）通过轨迹推算，累加运动矢量，得出当前位置的相对定位方法，单目里程计仅使用单个相机作为图像获取载体，使获得信息的要求更低，且能较精确地识别和定位特征点，实时性好，成本也少很多，因此具有更广的应用前景。本课题采用SURF算法来同时检测和匹配特征点，使用一种基于机器学习算法（SVM）自适应卡尔曼滤波器，减缓原本卡尔曼滤波器中会出现的精度低和发散状况，起到优化单目里程计的系统准确度。