Estimating the position and orientation, i.e. the pose of a vehicle in an unknown environment is even with known methods a challenge. For this reason, a map of the environment first needs to be created. The Simultaneous Localization and Mapping (SLAM) problem covers both challenges: determining the vehicle's pose as well as creating a map. The theory of SLAM is well studied and several approaches solving the problem already exist. A common approach is EKF-SLAM, which makes use of the Extended Kalman Filter (EKF). Another problem arising in the context of SLAM is the data association of sensor measurements with the map. For this purpose, the Nearest Neighbor Standard Filter (NNSF) approach is used. This is a well-known approach, which only considers likely associations and accepts the most probable among them. This work focuses on feature-based maps, which means that outstanding features in the environment - in our case visual markers similar to QR codes on consumer products - are used for localization and mapping. The aim of this thesis is to implement known algorithms for feature-based EKF-SLAM. On the one hand, this covers exploiting the benefits of the visual markers' IDs for known correspondences between detected visual markers and map features. On the other hand, the EKF-SLAM result is improved by observations without IDs using NNSF resolving the unknown correspondences. This yields a hybrid approach for the data association problem. The scientific contribution is a measurement noise model for the visual marker detection used. For this purpose, measurements of visual markers are recorded and statistically evaluated in order to derive approximation functions for the measurement variance. Finally, the obtained measurement noise model and the discussed algorithms are put into practice by providing a package for the Robot Operating System (ROS). The resulting implementation has been shown to be applicable to a simulated environment.