Semi-automatic sign board reconstruction from a land vehicle mobile mapping system

There is lots of semantic information in the sign board related to its corresponding building. For example, we can realize the property of the shop via the sign board. Previous study on sign board detection and recognition from mobile mapping system (MMS) is rare. However, if this process could be done automatically, it will increase the efficiency for POI (point of interest) collection. In this paper, we capture the street view images from a land vehicle MMS, which was configured with three stereo cameras by viewing the front of car and two street sides. At first, we detect straight line-segments by an open source called Line Segmentation Detector (LSD) and perform semi-automatic line-segment matching, then using line-plane intersection algorithm for space intersection to obtain the sign board boundary on the object space. Finally, we perform geometrical reconstruction to get 3D model of the sign board, based on the assumption that the sign board is made in a thin plate. In this study, some factors affect the line segment detection will be analyzed, such as shadow, occlusion, and shape of the sign board, image resolution, and geometric distortion introduced by cameras' viewing angles. Experimental results show that the above mentioned effects will decline the successful rate; particularly the shape of sign board must be complete and clear. Meanwhile, the scale change within the image will introduce spatial resolution variation, thus further studies are necessary to cope with those deficiencies in order to increase the success rate, reliability and robustness. Nevertheless, two successful examples will be illustrated to demonstrate its potential in POI collection.