Research Of Object Detection And Tracking For Augmented Reality

Object detection and tracking is a basic technology in the field of augmented reality. The traditional algorithms of object detection and tracking apply an affine transformation to a uniformly distorted image, instead of the projective transformation. The object detection and tracking technology for augmented reality is devised for estimating a general projective transformation between the object image and video images which contain the object. We calculate the position of the object in the real scene to ensure that the virtual objects are superimposed on the real scene images in the proper positions. Consequently, this paper puts emphasis on the projective transformation.In the traditional methods, the object detection and tracking method based on SIFT feature matching algorithm is one of the most commonly used algorithm due to its scale and rotation invariant features. But the time complexity of a traditional SIFT feature matching algorithm is so high that it is difficult to achieve real-time performance in an augmented reality system. This paper presents a real-time object detection and tracking method based on the improved SIFT feature matching algorithm. Firstly, for rapid and affordable feature extraction, we use AGAST corner detection instead of DoG scale space extrema detection. Then the number of descriptor dimensions is reduced from128to36. Finally we build multiple random K-d tree for feature matching and estimate the homography transformation parameter by progressive sample consensus algorithm.Motion blur may be introduced by a great relative motion between a camera and the object which could lead to SIFT feature matching algorithm failure. Based on projective motion blur model, we propose an object tracking strategy under motion blur. Experimental results show that the object detection and tracking algorithm presented in this paper has good real-time performance and robustness, and improved object detection and tracking success rate.