面向动态环境的机器人同步定位与建图技术

Research on robot simultaneous localization and mapping technology in a dynamic environment

传统的同步定位与建图(simultaneous localizition and mapping) SLAM算法采用了静态世界的强假设,这个假设限制了多数视觉SLAM算法在真实环境下的应用。针对这一问题,提出一种基于RGB-D传感器的动态SLAM算法。该方法是基于ORBSLAM2的改进,能够一致地映射包含多个动态元素的场景,并且增加了动态对象语义分割的能力。该方法结合深度语义网络与几何分割方法,对环境中的动态物体进行检测与分割,并去除对应的特征点,减少了动态物体的影响。TUM-RGB-D动态数据集序列中的实验结果表明,本文提出的系统大大提升了ORBSLAM2在动态环境下的定位精度,并且与其他先进的动态SLAM系统相比,精度有了一定程度的提升。

Traditional SLAM algorithms adopt the typical assumption of static scenarios,which limits the application of most visual SLAM(simultaneous localizition and mapping)algorithms in a real environment.In order to solve this problem,a dynamic SLAM algorithm is proposed based on RGB-D sensor.This method is based on the improvement of ORBSLAM2.It can consistently map scenes containing multiple dynamic elements and add the capabilities of the semantic segmentation of dynamic objects.It combines deep semantic network and geometric segmentation method to detect and segment dynamic objects in the environment.It removes the corresponding feature points to reduce the impact of dynamic objects.The experimental results in the dynamic data set sequence of TUM-RGB-D show that the method proposed in this paper greatly improves the localization accuracy of ORBSLAM2 in a dynamic environment,and its accuracy has been improved to a certain extent compared with those of other advanced dynamic SLAM methods.