Augmented Reality System for Accelerometer Equipped Mobile Devices

Abstract

In this paper, we present a novel approach for mobile augmented reality system. We estimate the 3D camera pose by detecting local invariant image features and combining them with the camera’s accelerometer data. We applied NELFD - Neuroevolved Local Feature Descriptor that encodes data around points of interest in the image using a neural network with evolved topology and weights. For every image frame, a correspondence between 2D feature points is calculated and the camera’s pose is established based on additional sensor information. Generally mobile systems are low performance and equipped with low-grade camera. Thus, due to estimation accuracy and low computational complexity our approach has been considered as a new alternative in the mobile augmenting process. Experimental evaluation proved that our method is capable of real-time pose tracking and augmentation in an unconstrained environment.