抄録
As the practical uses of point cloud generated by the Mobile Mapping System (MMS) spread, modeled data are applied to many fields. However, the cost of modeling process increases with the size of data. So, in order to make the process efficient, we have proposed recognizing pole-like objects method. The proposed method recognizes pole-like objects focusing on local area of segments and using Support vector Machine (SVM). The method recognized pole-like objects robustly and its recognition accuracy is 91.4%. However, this method recognized walls as pole-like objects (miss recognition) and wasn't able to recognize tilt pole-like objects. Thus, we added two improvements to the method. One is tilt correcting in machine learning and recognition phase. The other is making local area search active. With these improvements, the total recognition accuracy became 97.4%.
