Markerless augmented reality (MAR) aims to integrate 3D virtual objects into the real environment in real-time, enhancing the user's perception and interaction. It differs from marker based augmented reality (AR) systems basically by the method used to place virtual objects in the real world. In MAR, the real environment may be used as a marker that can be tracked in order to position virtual objects. Generally, the techniques used are based on computer vision, image processing, and computer graphics, and a major issue related to the field is robust, yet precise real-time object tracking and registration. Several AR solutions using general purpose devices have been developed. Such processing is done by software, making it difficult to obtain real-time results without compromising resolution and frame rate, and requiring the use of high clock frequencies that consequently lead to higher costs and power consumption. This paper introduces MARCam, an FPGA based solution that allows the development of embedded MAR applications. This framework allows the development of compact hardware with wearable capabilities for applications requiring user mobility through unknown environments and real-time dedicated processing. One of the most promising applications of this technology is guidance in training systems. MARCam can utilize a Structure from Motion (SfM) based technique. Instead of relying on previously obtained information about the real scene, SfM based techniques estimate the camera displacement without a priori knowledge of the environment. These methods are also able to retrieve the structure of the scene in real-time, with various levels of detail. Due to this, it is possible to reconstruct a totally unknown environment on the fly. MARCam's architecture is divided into many circuit modules, each one responsible for a specific task. This way, one can quickly arrange an assembly of modules and have in short time a fully-working, dedicated MAR system.