Augmented reality (AR) has been widely used in the field of education, which puts forward higher performance for AR systems. The extraction of feature is the basis of implementing tracking registration in AR systems. The latest feature extraction methods cannot well meet the needs in the education field. Therefore, we propose a new feature extraction algorithm: On the basis of extracting the edge of the image, the star-shaped projection values are calculated for all the pixels in the detection area for initial extraction of features. Then the false features are gradually eliminated by judging the main peak areas and the distance between two main peaks. Finally, binary robust independent elementary features (BRIEF) descriptor is used to describe features and hamming distance is used to match features. The experimental results show that, the average time of feature matching in this paper is only 22% of the improved speed up robust features (SURF) algorithm, and 87% of the improved oriented rotated BRIEF (ORB) algorithm. The correct matching rate of the features is higher than both improved SURF and improved ORB in the case of illumination variation, scale variation and visual angle variation.