Multi-scale Features and Markov Random Field Model for Powerline Scene Classification

Timely and accurate monitoring the safety of power line can prevent dangerous situations effectively. It is proposed that a Markov random field(MRF) model, into which a random forest classifier being integrated, to classify airborne LiDAR point cloud for power line scene. First, it is extracted that multi-scale visual features according to spatial pyramid theory to represent geometry information of the point and its neighborhood. And then a random forest classifier is used to describe the probability distribution of observed data. Meanwhile, contextual prior probability is established using MRF model, which is formulated as a multi-label energy function. Finally, the multi-label graph-cut technique is used to minimize energy function for optimizing the labels. It is validated the proposed method with LiDAR point cloud acquired by helicopter and mini-UAV power line inspection system. Experimental results demonstrate that the model can effectively classify pylon, power line and vegetation, with the overall accuracy of over 98%. Moreover, compared with other methods, the proposed model shows higher classification accuracy, particularly for the classification of the pylon