Robust Identification of Visual Markers Under Boundary Occlusion Condition*

Abstract

Visual markers are widely used in indoor marker-based positioning systems to achieve higher speed and accurate positioning performance. However, the classic mark identification methods have certain limitations when encountering complex conditions. Especially, if the marker’s boundary is blocked, it is almost impossible to identify the marker. In this paper, we redefine the identification task to a classification task based on CNN method. We also do some image transformations to create the dataset for our task. We train our dataset by transfer learning based on Google’s Inception-V3 CNN model. The experimental results show that the classification method can handle the boundary occlusion problem well, which is also proved to be useful for other complex conditions.