基于激光视觉融合的帧间匹配方法

Frame-matching Method Based on Laser and Vision Fusion

同步定位与地图构建(Simultaneous Location and Mapping,SLAM)是机器人在未知环境中实现自我导航能力的重要保证。目前SLAM算法使用的主传感器基本是激光雷达或视觉相机。二者各具优劣,激光雷达能更精确地进行测距,视觉相机能反映环境丰富的纹理信息。与使用单一传感器相比,将二者融合的SLAM算法能够获得更多环境信息,达到更好的定位和建图效果。文章提出一种融合激光雷达和视觉相机的帧间匹配方法,通过在SLAM帧间匹配过程中加入地面约束以及视觉特征约束,提高帧间匹配过程精度,增强算法鲁棒性,从而提升SLAM算法整体效果。文章最后利用采集的地下停车库数据进行结果验证,与开源算法A-LOAM进行对比。结果表明,相比A-LOAM的帧间匹配方法,文章提出的方法相对位姿误差提升约30%。

Simultaneous location and mapping(SLAM)is an important guarantee for robot to realize self navigation ability in unknown environment.At present,the main sensors used in SLAM algorithm are lidar or vision camera.Lidar can range more accurately,and visual camera can reflect the rich texture information of the environment.Compared with using a single sensor,SLAM algorithm can obtain more environmental information and achieve better positioning and mapping effect.In this paper,an inter frame matching method combining lidar and vision camera is proposed.By adding ground constraints and visual feature constraints in the process of slam inter frame matching,the accuracy of inter frame matching process is improved,the robustness of algorithm is enhanced,and the overall effect of SLAM algorithm is improved.Finally,the paper uses the collected underground parking data to verify the results,and compares with the open source algorithm A-LOAM.The results show that the relative pose error of the proposed method is improved by about 30%compared with A-LOAM.