【摘要】 已有的基于视觉的SLAM方法大多采用双目立体视觉,存在成本高,标定过程复杂,鲁棒性低等缺点。因此,提出了基于单目视觉和里程计的SLAM方法。该方法采用尺度不变特征变换算法提取特征,用扩展卡尔曼滤波更新地图。此外,由于单个CCD摄像头不能直接获得图像的深度信息,采用特征点延迟初始化的方法来解决这个问题,这种方法使得基于单目视觉的导航变得可能。在MATLAB环境下实现了SLAM算法仿真,实验结果表明,在室内环境下,该算法运行可靠,定位精度高。更多还原

【Abstract】 The SLAM algorithm is mostly based on stereo vision that has some disadvantages of high cost, complexity of calibration and low robustness. Therefore,the monocular vision and odometer based method is proposed to solve this problem.This method uses the scale invariant feature transform algorithm to extract features from images obtained from different view point, then uses extended kalman filter to update the map.The depth information can not be extracted by a sigle camera. The problem can be solved by the landmarks delayed initialization method which makes the monocular vision based navigation possible.The simulation of the SLAM algorithm is implemented in MATLAB, the experiment result shows that the proposed method is feasible,and with high localization precision in indoor environments.更多还原