Long-term 3D Registration Method Based on LCT Tracking and Improved ORB Detection

Aiming at the complex environment such as fast moving of registered area, occlusion, illumination change and the high requirement of real-time and precision of feature detection in the 3D registration of augmented reality system, a long-term 3D registration method based on LCT tracking and improved ORB detection is proposed in this paper. Firstly, the reliability of LCT algorithm in long-term tracking is used to track the area to be registered in augmented reality; secondly, ORB algorithm with excellent real-time performance is improved by setting adaptive thresholds, number of feature points and distance thresholds to optimize the dense area of image feature points. Parallel algorithm is used to retain feature points with larger eigenvalues, and discrete difference feature is used to enhance illumination unevenness. The stability of ORB operator under uniform change can solve the problem of low precision of feature detection and poor anti-jamming ability. Finally, the 3D registration matrix is calculated by using the detected feature points to enhance the real world. The simulation results show that the LCT algorithm has high reliability in long-term tracking and registration. Compared with ORB algorithm, the improved ORB algorithm improves the precision of feature detection by about 22%. It effectively improves the real-time and precision of feature matching in augmented reality system. The performance of the long-term 3D registration method based on LCT tracking and improved ORB detection is excellent, which improves the robustness, stability and practicability of the augmented reality system.