基于轨迹拐点滤波的激光雷达里程计定位算法研究

Research on LiDAR Odometry Localization Algorithm by Filtering Based on Inflection Points of the Trajectory

关键词:  权重模型  粒子滤波  拐点  OSM  激光雷达里程计

Key words:  Weight model  Particle filter  Inflection point  OSM  LiDAR Odometry

摘要:

在GPS拒止环境下，激光雷达里程计可以利用帧间匹配跟踪车辆实现定位，但是定位误差随时间累积的特性造成激光雷达里程计(LO)缺乏持续性。为解决LO的误差累积问题，引入轻量级地图OSM。基于粒子滤波框架，以LO作为运动模型的输入，通过两次筛选提取拐点，利用拐点匹配完成与地图的对齐，并以粒子的均值作为车辆校正后的位置，实现对定位误差的校正。提出了一种新的粒子权重模型，利用不同节点的相似度模型及测量值作为粒子权重的更新依据，避免拐点与路网节点的错误关联导致定位误差加大。经由KITTI数据集上的实验验证，该算法可以有效克服LO误差漂移问题，相较于原始LO定位精度至少提高了49.22%，且具有较好的实时性。

Abstract:

In a GPS-denied environment, LiDAR Odometry(LO) can utilize frame-to-frame matching to track vehicles to achieve localization, but it's impossible to use LO for a long time because of its error accumulation over time. To solve the error accumulation problem of LO, a lightweight map called OSM is introduced. Based on the particle filter framework, LO is used as the input of the motion model, the inflection points are extracted through two screenings, the inflection point matching is applied to complete the alignment with the map, and the mean value of the particles is chosen as the corrected position of the vehicle to realize the error correction of localization. A new particle weight model is proposed, and the similarity models and measurements of different nodes are used as the update basis of particle weight. In this way, the wrong association between the inflection points and the road network nodes, which may lead to an increase of localization error, can be avoided. By testing on the KITTI dataset, it is verified that the algorithm can effectively overcome the LO error drift problem, and the localization accuracy is improved by at least 49.22% compared with the original LO, with better real-time performance.