Research On Image Local Feature Detection Method And Its Application In Mobile Augmented Reality

The image feature detection is the basis of solving many problems in the field of computer vision, including object or scene recognition, 3D structure restoration of multiple images,consistency search of the dimensional space, and motion tracking etc. The local feature detection of image is an important part of the image feature detection. In recent years, because of it has a strong robustness for blur, light, and rotation of image and so, which had paid much attention.Due to the limitations of mobile platform, the image feature detection algorithms of PC platform were still unable to satisfy real-time and robustness requirements in the mobile platform.In this paper, the existing methods had been made some improvements according to the characteristics of the mobile platform, and the improved method was applied to a mobile augmented reality system based on the iOS platform, and it had achieved good results. The specific work and main results of this paper are as follows:Firstly, an improved fast SURF local feature detection method of image was proposed.Aiming at the problem that the traditional SURF feature detection method can't satisfy the real-time processing on the mobile platform, a new method which combined FAST feature detection and SURF feature description was proposed. Firstly, the original image is smoothed and sharpened to improve the stability of the feature points in the image. Then, the feature points are detected by the FAST method which ensures the fastness of the feature point detection process,and that are described by the SURF descriptor which ensures the robustness of the feature points.The experiments show that the proposed method can greatly improve the time efficiency of the algorithm and better satisfy the real-time requirements of the mobile platform but a slight decrease in the accuracy rate.Secondly, a multi-scale FAST&SURF image feature detection method was proposed. In order to overcome the problem of missing the scale information of the classical FAST detection method, a multi-scale FAST feature detection method was proposed, and which combined with SURF feature description. Firstly, the image pyramid is constructed, and the classical FAST feature detection method is extend to the scale space, then the feature point strength is determined by the defined feature response function, the feature points with the position and scale information are obtained by the extreme search. Finally, the feature points are described by the classical SURF descriptor. The experiments show that the algorithm has the expected effect in terms of time efficiency and robustness.Thirdly, the mobile augmented reality system based on the iOS platform was designed and implemented. The proposed multi-scale FAST&SURF algorithm in this paper was used in this system to match the feature point detection of the scene and the target object and handle the real-time tracking. The implementation of the system further verifies the feasibility and practicability of the proposed method.