A Study on the Feature Matching Performance of Randomized Trees

In order to improve the augmented reality systems, is presented an evaluation of the randomized tree algorithm for augmented reality feature matching, and compared it with the scale-invariant feature transform algorithm. Tests including the adaptability and matching speed on rotation, scale and illumination changes were carried out. Experimental results showed that random tree can achieve 30 frames per second of real-time feature matching, and moreover, can reach more than 50% of the matching rate when illumination changes, however its matching accuracy still needs to be improved.