Localization is essential for construction machines navigation. Current construction machines use RTK-GNSS for this purpose. We consider that using on-board cameras with SLAM algorithms to estimate machines' ego-motion and position change would not only be cheaper, but also more robust and widely applicable. To investigate the feasibility of this idea, we installed several consumer cameras on real construction machines and performed experiments at real construction sites. Our experiments show promising results, achieving localization accuracy of about 0.06 meters.