An object recognition approach based on structural feature for cluttered indoor scene

In this paper, aiming at the indoor scene under monitoring by visual sensor network (VSN), an object recognition approach based on structural feature is presented. Firstly, we regard the output of existing line segment detector LSD with proper parameters as the preliminary extraction result and it still will be further restored and split. Then, we give an inference model based on structural features of object including line segment ontology characteristics and relative relationship between the line segments. Finally, the objects are recognized with position information through inference. The effectiveness of the approach is verified, and the results show that our approach does not rely on segmentation and has robustness on partial defect and structural deformation to some extent.