Simultaneous Localization and Mapping (SLAM) algorithms are a vital component of a mobile robot system in order for it to create a map and navigate through its environment. Depending on the complexity and size of the surroundings, different sensors and SLAM alternatives may be utilized. Laser rangefinders are traditionally popular sensor options, but their weakness is the restriction of measuring object distance only in a single plain. Therefore, laser vision cannot detect essential information out of their the field of view, which can lead to a collision.

A complex environment is a space consisting of obstacles in various forms and sizes. In this work, the complex environment problem (the challenge of safely traveling through the environment) is approached with an RGB-D camera, which can produce a 3D point cloud of the scene. The point cloud is then converted into a 2D laser format to execute a particle filter based SLAM algorithm and create an occupancy map. This thesis studies the effect of restricting the height of the point cloud on the SLAM process so that the system considers the physical size of the mobile robot. Data sets were recorded in an office environment with different sizes and forms of obstacles positioned in the area. The evaluation considers five different elements of the system, which are mostly related to the SLAM process. The analysis includes visual evaluation of maps, resampling, and scan matching process (demonstrating the effect on SLAM), CPU load, and Root Mean Square Error(RMSE) for accuracy analysis.

Results show that the point cloud could be restricted to the height of the mobile robot without a notable compromise on accuracy. It is also shown that a significant point cloud restriction before conversion into 2D laser, will increase scan matching failures and therefore affect the SLAM process.