Intelligent mobile robots leverage visual odometry (or SLAM, simultaneous localization and mapping) techniques to achieve localization and planning tasks in environments devoid of a priori map information. Employing multiple sensors has the potential to compensate for individual shortcomings, thereby enhancing the localization performance of the system, while introducing added complexity to system design. In this paper, we propose the visual odometry utilizing multiple monocular cameras with non-overlapping FoV (Field of View) for a mobile robot constrained to move in a two-dimensional horizontal plane. Our approach maximizes the observation of each monocular camera, addresses the challenge of up to scale inherent in monocular methods, and introduces an optimization strategy for online extrinsic calibration tailored for motion-constrained scenarios.