Designing a Simple Fiducial Marker for Localization in Spatial Scenes Using Neural Networks

Abstract

The paper describes the process of designing a simple fiducial marker. The marker is meant for use in augmented reality applications. Unlike other systems, it does not encode any information, but it can be used for obtaining the position, rotation, relative size, and projective transformation. Also, the system works well with motion blur and is resistant to the marker’s imperfections, which could theoretically be drawn only by hand. Previous systems put constraints on colors that need to be used to form the marker. The proposed system works with any saturated color, leading to better blending with the surrounding environment. The marker’s final shape is a rectangular area of a solid color with three lines of a different color going from the center to three corners of the rectangle. Precise detection can be achieved using neural networks, given that the training set is very varied and well designed. A detailed literature review was performed, and no such system was found. Therefore, the proposed design is novel for localization in the spatial scene. The testing proved that the system works well both indoor and outdoor, and the detections are precise.