基于平面约束的RGB-D SLAM系统

RGB-D SLAM system based on plane constraint

针对现有室内环境存在多个平面特征的特点,提出了一种使用平面特征优化相机位姿和建图的实时室内RGB-D同时定位与地图构建(SLAM)系统。系统在前端采用迭代最近点(ICP)算法和直接法联合估计相机位姿,在后端提取相机关键帧平面特征并建立多种基于平面特征的约束关系,优化相机关键帧的位姿和平面特征参数,同时增量式地构建环境的平面结构模型。在多个公开数据序列上的实验结果表明,在平面信息丰富的环境中,平面特征约束能够减小位姿估计的累积误差,系统能够构建场景平面模型并只消耗少量的存储空间,在真实环境下的实验结果验证了系统在室内增强现实领域具有可行性和应用价值。

Considering that there are many planar features in the indoor environment,this paper proposed a real time indoor RGB-D simultaneous localization and mapping(SLAM) system using planar features to optimize camera poses and map.In frontend,this system employed iterative closest point(ICP) and direct method to jointly estimate camera poses.In backend,this system extracted planar features of camera keyframes,and constructed multiple constraints based on planar features to optimize camera keyframes poses and parameters of planar features,and incrementally constructed planar structure of environment.The results of experiments on multiple dataset sequences show that planar feature constraints can reduce the cumulative estimation errors of poses and this system can construct planar structure of environment with less storage space,the results of experiments on real environments show that system has feasibility and application value in the field of indoor augmented reality.