Maintaining stable and accurate registration is the key to build augmented reality（ AR） system

in order to improve stability and precision of registration

a multi-sensor fusion registration algorithm based on complementary filtering（ CF） and unscented Kalman filtering（ UKF） is proposed. This algorithm adopts complementary filter is used to estimation the attitude of camera by fusing datas of gyroscope

accelerometer and geomagnetic sensor

features are matched by combining intertial attitude datas when visual images are processing

UKF is used to estimate position by fusing visual and inertial datas. Experimental results indicate that the tracking stability

precision

efficiency and anti-interference ability of this algorithm are superior to traditional fusion algorithm based on extended Kalman filtering（ EKF）.