Ground robot multi-scale road perception based on semantic tree MRF model

Road real-time perception is the key technology of autonomous ground mobile robot to realize autonomous navigation. But it is difficult to develop a road sensing algorithm because of the complexity and uncertainty of outdoor road environment. A multi-scale biomimetic road sensing algorithm in wavelet domain based on semantic tree Markov model is proposed. In time-space domain, the three-dimensional random field is used to express the road image sequence collected by robot. A road model named road best tree-Markov random field (RT-MRF) using the semantic tree structure Markov random field (MRF) is proposed. The genetic algorithm is used to optimize the supervised RT-MRF model for image segmentation of road sequences. The road recognition and autonomous navigation are realized through tracking segmentation boundary. An independently developed quadruped bionic robot is used as the research and experiment platform. The experimental results show that the proposed algorithm is a robust road image sequence segmentation method, which can be used under the poor detection conditions, such as shadow, cracks, holes, uneven and illumination change. And the real time of the algorithm is enough high to meet the demand of outdoor mobile robot autonomous navigation.