Hybrid tracking for outdoor augmented reality system

Till now there is not a sensor that can provide a complete solution for tracking in outdoor augmented reality system. To improve the robustness and accuracy of real-time visual tracking, we present a sensor fusion algorithm by combining inertial sensors and a CMOS camera, making it suitable for unknown environment. The fusion algorithm makes use of an extended Kalman filtering to fuse inertial and vision data to estimate the trajectory of the camera. Meanwhile, the inherent error drift problem of the inertial sensor is addressed by using the vision information. The method of single-constraint-at-a-time (SCAAT) is also introduced to assimilate the sequential observations. Experimental results show that the proper use of additional inertial information can effectively enhance the robustness and accuracy of visual tracking.