基于点线结合特征的单目视觉里程计

Monocular Visual Odometry Based on Point and Line Features

SLAM(即时定位与地图构建)系统是近年来计算机视觉领域的一大重要课题,其中特征法的SLAM凭借稳定性好、计算效率高的优点成为SLAM算法的主流。目前特征法SLAM主要基于点特征进行。针对基于点特征的视觉里程计依赖于数据质量,相机运动过快时容易跟丢,且生成的特征地图不包含场景结构信息等缺点,提出了一种基于点线结合特征的优化算法。相较于传统基于线段端点的六参数表达方式,算法采用一种四参数的方式表示空间直线,并使用点线特征进行联合图优化估计相机位姿。使用公开数据集和自采集鱼眼影像数据分别进行实验的结果表明,与仅使用点特征的方法相比,该方法可有效改善因相机运动过快产生的跟丢问题,增加轨迹长度,提升位姿估计精度,且生成的稀疏特征地图更能反映场景结构特征。

In recent years,SLAM has been an important topic in the field of computer vision.is one of the essential tasks in computer vision area.Among all the algorithms,feature-based SLAM stands out for its robustness and efficiency,especially point feature-based SLAM.However,as the visual odometry based on point features depends on data quality and is difficult to be tracked when the camera moves too fast,and the map constructed contains little scene structure information,a method based on point and line combination features is proposed.Rather than the traditional six-parameter representation of two end points for lines,the proposed algorithm applies a four-parameter representation to express space lines,and uses both point and line features to optimize camera position via graph optimization.Experimental results on public datasets and self-collected fisheye camera image sequences show that compared with methods that only use point features,the proposed algorithm can effectively make improvements on lost tracking caused by camera moving too fast,hence increasing the length of trajectory as well as the accuracy of position estimation.The structure of the scene is also better represented in sparse point-line feature map.