A method and apparatus of tracking poses of a rolling-shutter camera in an augmented reality (AR) system is provided. The method and apparatus use camera information and inertial sensor readings from Inertial Measurement Unit (IMU) to estimate the pose of the camera at a reference line. Thereafter, relative pose changes at scanlines may be calculated using the inertial sensor data. The estimated reference pose of the camera is then further refined based on the visual information from the camera, the relative pose changes and the optimized reference line pose of a previous image. Thereafter, the estimate of the scanline poses may be updated using the relative pose changes obtained in the earlier steps.