Practical Image-Enhanced LBS for AR Applications.

We have designed a multisensor indoor LBS suitable for augmented reality applications which, mainly based on computer vision techniques, provides precise estimations of both the 3D position and rotation of the device. Our proposal makes use of state-of-the-art IMU data processing techniques during the training phase in order to reliably generate a 3D model of the targeted environment, thus solving typical scalability issues related to visually repetitive structures in large indoor scenarios. A very efficient camera resection technique will then be used in the on-line phase, able to provide accurate 6 degrees of freedom estimations of the device position, with mean errors in the order of 5 cm and response times below 250 ms.