Research On Tracking Techniques For Mobile Augmented Reality

Augmented Reality (AR for short) is also known as mixed reality. It merges the real world and virtual data through graphics and other computer technology. Besides, it augments the real world to help transmit information and enhance the entertainment effects. With the popularity of notebooks, smart phones, PDAs and other mobile devices, people increasingly rely on these portable yet powerful guides or helpers. On the other hand, the AR applications on mobile platforms are widely accepted and welcomed by the public, due to its approaching to life, convenience and efficiency. Tracking technology has always been the core technology in the field of computer vision. In recent years, the rapid development of AR has made new application prospects of the tracking technology.The main purpose of this article is to explore and study the tracking technology in the mobile AR System. We will focus more on the less computational, high-performance algorithm, due to the limitation of resources and processing capacity on mobile platforms. As to say, we maximized the accuracy and stability while ensuring the speed. In this arti-cle, including camera tracking and object tracking, specifically designed and implemented a SLAM system on mobile phone and an object detection system on mobile computer system. In addition, the article analyzed and introduced the feature point detection de-scription and trajectory tracking from the algorithm point of view. It achieved some of the traditional algorithms, and proposed some new solutions.Firstly, we proposed a new feature detection algorithm called SE-FAST. In the ob-ject tracking and detection process, the biggest challenge was the Image Abstraction on the target objects and random scenes, while the core of Image Abstraction is feature extraction. Our SE-FAST was an algorithm of feature point extraction in the mobile AR application. SE-FAST established Characteristic response functions for traditional FAST, and greatly enhanced the quality of traditional FAST detection through regional extremum search. The FAST algorithm which was based on characteristic response con-tinued to expand to the scale-space. So that it gained scale invariance as well as scale data to better guide the subsequent description of algorithm and greatly increase the accuracy of feature matching. We also tried to optimize the basic detection algorithm of traditional FAST, to further improve the detection quality.Secondly, we established a SLAM system on mobile phone on the basis of single SE-FAST. We integrated the traditional SURF descriptor and calculated vision-based matrix with 8-point method, to complete the feature point detection as well as the establishment and updating of 3D localization and mapping diagram. Then, we created an object detection system on notebooks, which generated feature vectors and matched relevant data through the improved SURF descriptor. The final detection system was able to detect target objects in a complex natural environment. Meanwhile, the object pose, image location and size were given. The whole detection system was proved to own good computing performance and excellent robustness of scale, rotation and perspective change.Finally, we experimentally compared our SE-FAST and the current fastest detectors called Fast-Hessian. We showed efficient processing and high-quality detection results of SE-FAST. We also compared the improved FAST algorithm with the traditional one and the one based on characteristic response. It showed that the characteristic-response-based one and the improved one, especially the latter, had high-quality detection results, the detection rate of which was comparable to the traditional one. We also gave the experimental data of SLAM system on mobile phone and object detection system on notebook, to analyze the result and prospected for the future works.