为提高移动机器人的跟随精度,对深度相机(RGB-D相机)测距进行研究,提出一种基于MRSD(Mask R-CNN and S2R-DepthNet)的移动机器人跟随系统.引入实例分割算法(Mask R-CNN)获取行人的前景掩膜;以掩膜为指导从深度图像中获取准确的行人区域深度像素,引入深度估计算法(S2R-DepthNet)从彩色图像中推理深度图像以替换深度传感器引起的无效深度像素,提高测距的精度;建立基于Sage-Husa自适应滤波(SHAKF)的测距模型,提高量测信息异常情况下的测距鲁棒性,实现稳定跟随.实验结果表明,该方法能以设定距离准确跟随前方行人.
Abstract
To improve the following accuracy of mobile robots,the depth camera(RGB-D)ranging was investigated and a mobile robot following system based on MRSD(Mask R-CNN and S2R-DepthNet)was proposed.The Mask R-CNN was introduced to extract pedestrians'foreground masks.The masks were used as guides to extract pedestrians'depth information from the corre-sponding depth images.The S2R-DepthNet was introduced to reason depth information from RGB images to replace the invalid depth pixels caused by the depth sensor,thus improving the ranging accuracy robustness.A ranging model based on Sage-Husa adaptive filter(SHAKF)was established to improve the ranging robustness in the case of abnormal measurement information,so as to achieve stable following.Experimental results show that the method can accurately follow the pedestrians in front with the set distance.
维普
万方数据