Visual-inertial odometry (VIO) for autonomous vehicles provides ego-motion estimates with the help of an inertial measurement unit (IMU). However, VIO in large-scale outdoor environments is limited in its ability to estimate translational motion owing to forward motion degeneracy. In this paper, we propose an approach to estimate the ego-motion of a vehicle using a camera, an IMU, and a speedometer. The speed measurement model is incorporated into the Bayesian VIO framework, and a state re-initialization is applied based on the speed measurements. Experiments using the public KITTI dataset show the superiority of the proposed method compared to conventional VIO and stereo-based visual odometry. Furthermore, the proposed method achieves about 20 Hz frame rates for real-time autonomous driving.