A novel cascaded observer design exploiting source points' bearing measurements is proposed for the monocular visual SLAM problem. A distinguishing feature of the present work with respect to most existing visual SLAM algorithms is the decoupling of the camera's pose estimation (i.e., localization) from source points' position estimation (i.e., map building), leading to a straightforward architecture that can handle a vast number of source points efficiently. Furthermore, the persistence of excitation of the camera's translational motion together with the source points' configuration (specified in our prior works) is the key to achieving (local) exponential stability of the camera's pose estimation and, subsequently, overcoming the wellknown depth ambiguity associated with the use of a monocular camera. This ingredient has paved the way for the proposed cascaded observer architecture, in which the main contributions concern the design and stability analysis of the three proposed observers for source points' position estimation. Convincing comparative simulation and experimental results are reported to support the proposed approach.