DeepMoVIPS: Visual indoor positioning using transfer learning

Finding the location of a mobile user is a classical and important problem in pervasive computing, because location provides a lot of information about the situation of a person from which adaptive computer systems can be created. While the inference of location outside buildings is possible with GPS or similar satellite systems, these are unavailable inside buildings. A large number of methods has been proposed to overcome this limitation and provide indoor location to mobile devices such as smartphones. With this paper, we propose a novel visual indoor positioning system DeepMoVIPS, which exploits the image classification power of deep convolutional neural networks for symbolic indoor geolocation. We further show, how to transfer visual features from deep learned networks to the application domain and give encouraging results of more than 95% classification accuracy for datasets modelling work environments using 16 rooms and evaluation over a time frame of four weeks.