基于快速视觉里程计和大回环局部优化模型的改进VSLAM算法

Improved VSLAM Algorithm Based on Fast Visual Odometry and Large Loop Local Optimization Model

从提高移动机器人视觉同时定位与地图创建(visual simultaneous localization and mapping,VSLAM)过程中的自主定位精度和实时性的角度出发,提出了一种基于快速视觉里程计和大回环局部优化模型的改进VSLAM算法.首先,在对特征点进行不确定性分析的基础上,对color GICP(color supported generalized iterative closest point)误差函数进行改进,采用帧到模型(frame-to-model)方式实现对数据集和模型集的快速配准,并结合卡尔曼滤波和加权方法实现对模型集的更新,提高位姿估计精度;其次,提出一种基于模型到模型(model-tomodel)配准的大回环局部优化模型,并结合g2o图优化方法对位姿估计累积误差进行快速局部优化,从而进一步提高自主定位精度和效率.数据集离线对比实验和实际场景在线实验均表明,本文方法不但有效提高了移动机器人VSLAM过程中的自主定位和建图精度,而且具有较好的实时性.

An improved visual simultaneous localization and mapping （VSLAM） algorithm based on fast visual odometry and large loop local optimization model is proposed in terms of the accuracy and real-time performance of autonomous localization in mobile robot VSLAM. First of all, the error function of color GICP （color supported generalized iterative closest point） is improved based on the uncertainty analysis on the feature points. Frame-to-model approach is utilized to achieve fast registration between data sets and model sets. And the model sets are updated through Kalman filtering and the weighting method. The accuracy of pose estimation is improved through the above steps. Secondly, a large local loop optimization model based on model-to-model registration is proposed and g2o is combined to optimize the accumulated error of the pose estimation quickly, which improves the accuracy and efficiency of autonomous localization further. The offline contrast experiments based on the datasets and the online experiments based on actual scenes show that, with the proposed algorithm, not only the accuracy of autonomous localization and map in mobile robot VSLAM are improved effectively, but also the real-time performance is guaranteed.