A monocular odometer for a quadrotor using a homography model and inertial cues

Abstract

Sate estimation using a monocular camera and an Inertial Measurement Unit (IMU) is considered in this paper. For the visual part, only frame-to-frame motion is considered which is easier to implement than methods involving tracking features over multiple frames. A homography model is developed, aided by the measurements of angular rates from on-board IMU. It is shown that given angular rates in two of the axes, the angular rate in the other axis can be recovered without ambiguity. In this paper, using a ventral camera, the yaw rate is computed from the homography matrix and is combined with IMU gyro rate to estimate yaw angle. The unscaled speed from the homography model is fused with acceleration from the IMU to estimate metric distance to the scene, the metric speed and acceleration. Finally, a visual odometry system is built. The effectiveness of the proposed method is proven using real images and IMU data from our quadrotor platform.