Date of this Version
It has been a challenging subject to recognize dynamic objects from a scattered work environment because large and complex 3D site data obtained by a laser scanner makes it difficult to process itself in real or near real time. This thesis introduces a model-based automatic object recognition and registration framework, Projection-Recognition-Projection (PRP), to assist heavy equipment operators in rapidly perceiving 3D working environment at dynamic construction sites. In this study, a digital camera and a hybrid laser scanner were used to rapidly recognize and register dynamic target objects in a 3D space by separating target object’s point cloud data from a background scene for a quick computing process. A smart scan data updating algorithm has been developed which only updates the dynamic target object’s point cloud data while keeping the previously scanned static work environments. Extracted target areas containing 3D point clouds were orthographically projected into a series of 2D planes with a rotation center located in the target’s vertical-middle line. Prepared 2D templates were compared to these 2D planes by extracting SURF (Speeded Up Robust Feature) features. Then, point cloud bundles of the target were recognized, and followed by the prepared CAD model’s registration to the templates. The field experimental results show that the proposed PRP framework is promising and can significantly improve heavy construction equipment operations and automated equipment control by rapid modeling dynamic target objects in a 3D view.
Adviser: Yong Cho