Scene Dedicated Feature Descriptor with Random Forest Training for Better Augmented Reality Registration

The most important part of an Augmented Reality system is the tracking system to support an accurate and robust registration. In outdoor environments, the continuously changing environmental characteristics and elements make hard to achieve this tracking process. The main point of this operation is that the descriptor has to work with great accuracy in all kind of situations. The most used descriptors have this distinctive capacity, but computers and mobile devices process them in a long time frame. This paper investigates a new trained, lighter, scene dedicated descriptor, which takes into account the scene characteristics. The descriptor is loaded with elements that can be computed faster and have distinctive information about the selected area. The complete descriptor is used for semantical feature extraction with the aid of a trained Random Forest classifier. For validation purposes, the descriptor was tested against the most used descriptors and in some cases it proved to be faster and equally reliable.