基于Bumblebee XB3的同时定位与地图构建算法研究

发布时间：2018-01-28 16:54

本文关键词： SURF EKF 双目视觉SLAM 3D信息　出处：《中国海洋大学》2015年硕士论文　论文类型：学位论文

【摘要】：准确的3D场景信息在自主驾驶和移动系统导航中是非常重要的因素。由于照相机成本低而且容易使用,智能机器人通常会根据从照相机获取的观测信息来作出决定。周围真实的环境是相当复杂的,观测也经常有噪声,或许还可能丢失信息。这些都使得Visual SLAM的实现比较困难。尽管在高速公路导航和简单环境的自主导航获得了成功的验证,在未知复杂的环境下的自主导航仍然是个没有解决的问题。本文研究的方法是从一系列的立体图像中估计机器人的位姿和准确的3d的环境特征。所有的数据都是来自Bumblebee XB3双目立体照相机,照相机是安装在智能小车的顶部。本文主要研究双目视觉SLAM系统,双目视觉SLAM主要集中在机器人位姿和环境特征的3D位置及其坐标。在估计机器人位姿时稀疏的特征被追踪和位置估计,特征集合太稀疏而不能准确的描述照相机拍摄的真实3D环境。从立体双目图像序列中获取准确的3D信息在计算机视觉和机器人自主导航中是一个核心步骤。我们运用循环SURF特征提取匹配算法来获得粗略的匹配特征,该算法计算量小,鲁棒性高。接下来运用RANSAC和对极几何约束消除误匹配点获得精确稳定的特征点。将得到的精确稳定的特征点输入到基于EKF的SLAM系统来估计机器人的运动路径。在室外非结构化环境下验证了双目视觉SLAM算法的准确性和有效性。为了准确描述真实环境,运用双目视觉SLAM算法的同时运用有效的立体匹配算法来获取浓密的特征点。按照估计出来的机器人路径,我们将这些浓密的特征点映射到机器人坐标系,最终得到基于点的3D重建地图。
[Abstract]:Accurate 3D scene information is a very important factor in autonomous driving and mobile system navigation because of the low cost and ease of use of the camera. Intelligent robots usually make decisions based on observations from cameras. The real environment around them is quite complex and the observations are often noisy. It may also be possible to lose information. All of this makes the implementation of Visual SLAM more difficult, although autonomous navigation in highway navigation and simple environments has been successfully validated. Autonomous navigation in unknown and complex environment is still an unsolved problem. The method of this paper is to estimate the pose of robot and accurate 3D environmental feature from a series of stereo images. All the data are. It's from Bumblebee. XB3 binocular stereo camera. The camera is installed on the top of the smart car. This paper mainly studies the binocular visual SLAM system. The binocular visual SLAM mainly focuses on the 3D position and coordinates of robot pose and environment features. The sparse features are tracked and estimated in the estimation of robot posture. The feature set is too sparse to accurately describe the real 3D environment taken by camera. Getting accurate 3D information from stereo binocular image sequences is a key step in computer vision and robot autonomous navigation. We use circular SURF feature extraction matching algorithm to obtain rough matching features. The computation of the algorithm is small. The robustness is high. Then the exact stable feature points are obtained by using RANSAC and the elimination of the mismatch points. The obtained exact stable feature points are input into the SLAM system based on EKF to estimate the exact stable feature points. The accuracy and validity of the binocular visual SLAM algorithm are verified in the outdoor unstructured environment. In order to describe the real environment accurately. The binocular vision SLAM algorithm is used to obtain dense feature points simultaneously. According to the estimated robot path, we map these dense feature points to the robot coordinate system. Finally, the point-based 3D reconstruction map is obtained.
【学位授予单位】：中国海洋大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.41;TP242

【相似文献】