Hybrid object tracking method for augmented reality systems using the Kalman filter

Most augmented reality solutions are based on computer vision techniques to detect and track objects using a camera and display a computer generated virtual content. However, main issues of such techniques are image transformations, chaotic environment and illumination level. In general, such noisy environment causes complex or even impossible task for object tracking in real-time. In case of computer vision tracking techniques, occlusions causes 3D virtual content to disappear, therefore it has negative impact for the usability of applications. Because of the named reasons, most widely used modern feature detection and description techniques were analyzed that allows to solve partial occlusion problems and properly display virtual content. In this research robustness comparison according to repeatability criteria was accomplished using different type of image transformations sets by verifying specific image pairs. Computer vision techniques performance estimates in respect of speed was also analyzed to inspect suitability for the augmented reality. Research results are critical to propose improved tacking techniques in augmented reality field, also, by integrating inertial sensor information. Quaternion application for microelectromechanical sensor data were also thoroughly analyzed to estimate camera orientation. With this solution it is aimed to eliminate separate sensor disadvantages and improve camera orientation tracking accuracy, reliability and virtual content representation in augmented reality, without using any computer vision technique. The proposed solution was adapted not only from tracking perspective, but also for the development of spherical form interaction device. Detailed Kalman filter analysis and practical application are also presented for orientation data acquired from real sensors, which are expressed in quaternions. Analysis of gradient descent method and practical application for sensor data fusion were also accomplished. The proposed hybrid tracking technique for augmented reality was proposed, where orientation and position estimates from computer vision and sensor fusion techniques were analyzed in a systemic approach. Sensor array was used to eliminate separate sensor disadvantages and improve tracking without using or partially using computer vision data. Provided research experiments has predefined assumptions and orientation-position data were imitated in respect to computer vision and sensor fusion. Additional conditions were introduced to acquire optimal orientation-position estimates and results were compared. Complementary properties of Kalman filter hybrid tracking technique were demonstrated. Solution was adapted for mobile devices in the field of augmented reality for internet of things. To achieve more accurate and reliable camera orientation and position estimates in space and expand augmented reality application possibilities, in this dissertation computer vision and sensor fusion technique efficiency were analyzed and a new hybrid object tracking technique using Kalman Filter was proposed. For this purpose. The main use of presented technique is to expand augmented reality application fields, also improve object orientation and position tracking for reliable virtual content representation even in cases, when useful information from feature extraction and matching from image is not available