一种优化的机器人室内场景定量分割识别方法

Optimization method for indoor scene quantitative segmentation of robots

为了改善机器人认知环境的能力,基于高效、简单、实用等设计目标,提出了一种通过物体表面凹凸度分割识别未知物体的方案。该方案以连续局部表面凹凸度取代布尔型判定,在结合表面凹凸度和法方向信息的基础上,提出一种新型的分割权重计算方法;该权重可以对需要分割的物体与场景交界、不能分割物体表面边缘进行标识。在此基础上为输入场景构造无向带权图,并通过快速图分割算法获取未知物体。经实验发现:相对于基于凹凸性的判定,基于凹凸度的量化衡量在应对观测噪声和估计误差上的鲁棒性更好。

In order to improve the robot’s cognitive environment,a scheme is proposed to detect unknown objects by means of the object surface roughness.The program replaces the Boolean determination to the continuous local surface irregularities,and on the combination of surface roughness and the information of the method,a new method for calculating the partition weight is proposed,which can reveal the boundary between the object and the scene and the edge of the object.On the basis of this,the construction of the input scene is constructed to the weighted graph,and the unknown object is acquired by the fast graph partitioning algorithm.The experimental results show that,compared with the method based on the convexity,the robustness of the method is better than that of the observation noise and the estimation error.