点线特征融合的双目视觉SLAM算法

Point-line Feature Fusion in Stereo Visual SLAM Algorithm

在低纹理场景中,基于点特征的同时定位与地图构建(Simultaneous Localization and Mapping,SLAM)算法很难追踪足够多的有效特征点,系统甚至无法正常工作.众所周知,丰富的线段特征存在在人造结构化环境中的地面与墙面交界处.因此,提出一种点线特征融合的双目视觉SLAM算法.在特征提取前,引入梯度密度滤波器加速线特征提取和提高线匹配的准确度,在特征点匹配阶段,采用渐进采样一致性(Progressive Sampling Consensus,PROSAC)算法剔除误匹配点,从而提高定位精度.此外,在特征的融合过程中引入加权思想.在构造误差函数时对点线特征权重进行合理分配.最后,通过在公开的数据集上得到的仿真并与一些优秀的算法进行对比,该算法性能优于PL-SLAM和LSD-SLAM算法,证明了算法的有效性和准确性.

In low-texture scenes,it is difficult to track enough effective feature points based on the simultaneous localization and mapping(SLAM)algorithm,and the system cannot even work properly.As we all know,there are abundant line segment features at the junction of the ground and the wall in the man-made structured environment.Therefore,a stereo visual SLAM algorithm based on point and line features fusion is proposed.Before feature extraction,a gradient density filter is introduced to speed up line feature extraction and improve the accuracy of line matching.During the process of feature point matching,the progressive sampling consensus(PROSAC)algorithm is introduced to eliminate the mismatched points and improve the positioning accuracy.In addition,the weighting idea is introduced in the process of feature fusion.When constructing the error function,the weights of point and line features are reasonably distributed.Finally,serveral experiments are introduced which compare the improved method with some state-of-art algorithms on the public datasets.It is proved that the performance of this algorithm is more effective than the PL-SLAM and LSD-SLAM algorithms,and it has higher accuracy.