Robust Feature Point Extraction and Tracking for Augmented Reality
For augmented reality applications, accurate estimation of the camera pose is required. An existing video-based markerless tracking system, developed at the ITS, presents several weaknesses towards this goal. In order to improve the video tracker, several feature point extraction and tracking techniques are compared in this project. The techniques with best performance in the framework of augmented reality are applied to the present system in order to prove systems enhancement. A new implementation for feature point extraction, based on Harris detector, has been chosen. It provides better performance than Lindeberg-based former implementation. Tracking implementation has been modified in three ways. Track continuation capabilities have been added to the system with satisfactory results. Moreover, the search region used for feature matching has been modified taking advantage from pose estimation feedback. The two former modifications working together succeed in improving systems performance. Two alternatives to photometric test, based on Gaussian derivatives and Gabor descriptors, have been implemented showing that they are not suitable for real-time augmented reality. Several improvements are proposed as a future work.