Performance analysis of markerless tracking systems for augmented reality applications in education

Technology is constantly evolving and brings innovations. At the beginning of these innovations are computers, mobile devices and the concept of Augmented Reality. The virtual image support of Augmented Reality technology, which provides visual enrichment of the environment by adding 2D or 3D virtual patterns such as sound, picture, video created in the computer environment to the real world, has enabled this technology to be used in many fields, especially in education. In addition, the high preference of computers and mobile devices today has enabled the rapid integration of Augmented Reality technology into these platforms. In this thesis, a markerless-based model, which takes natural attributes as a reference instead of marker, requires high computational power and memory capacity, was taken as reference, and it was aimed to evaluate the performance of the algorithms used in this direction and to evaluate the performance of the algorithms used in this direction, and then to integrate the appropriate algorithm into the android platform. In this study, the place of Augmented Reality technology in the literature, how it develops, its types and its place in the field of education are discussed and it is said that this technology is more efficient for students when used instead of classical teaching. However, it was seen that most of the application examples examined were developed with the marker-based model introduced to the system before. Since marker-based applications cannot provide marker independence, instead of adding markers to the system beforehand, within the scope of this thesis; corners, edges or shapes are referenced. The technologies required to use markerless-based model in Augmented Reality are examined and accordingly ORB, SIFT and SURF algorithms which require high computational power and memory capacity have been examined and performance tests have been carried out on the Netbeans platform. desktop and Android Studio platform for mobile devices. For the analysis phase, the feature detection, extraction, matching, tracking and running times were tested on a sample video for three different situations without brightness, color change and scaling. The data obtained as a result of the tests were supported by graphics and tables, and the performance analyzes were compared for each case. As a result of the study, it was seen that SIFT and SURF algorithms obtained more key points and feature vectors for each case compared to the ORB algorithm, therefore they made more matches. These matches were filtered using a certain ratio and good matches were found and better matches were found using the RANSAC algorithm and the deviation amount of the virtual object was tested for each algorithm. Tests have shown that the deviation rate is less when using the ORB algorithm and it is faster than SIFT and SURF in terms of time, especially for the mobile platform. In this direction, considering the processor power and memory capacity of the android, a simple application developed using OpenCV and OpenGL libraries has been integrated to introduce the 2D geometric shape to the education area in 3D platforms for Augmented Reality environment.