Using Deep learning for providing better AR experince

Two Turkish researcher have conducted a research on  Augmented Reality Applications Using Deep Learning and they provided valuable information

Unlike the totally artificial structure of the virtual reality, the augmented reality works by placing virtual objects on user’s vision while he/she watching the real world. In the first applications of augmented reality, a pre-defined pattern is used to do this, which is sufficient to place objects on the pattern. The position of the pattern in the camera was also giving direction to the object. It was a simple process of training the pattern and capturing it in the image. Afterwards owing to feature vectors used in systems which are trained by artificial neural networks, complex patterns can be used. Deep learning is a revolutionary technique in artificial intelligence applications. According to neural networks which are working with high error rates, this technique exceeds the recognition properties of the human brain. Instead of using a pattern, augmented reality applications that are use deep learning need to be educated an object which an information shown on it. For example, when the source object is a human hand, recognition can be performed in all of the different positions of any human hand. It is clear that deep learning will be effective in the future of the augmented reality.
You can read their paper at reasearchgate