Laser scanning devices enable the capturing and storage of scenes and objects in the real world. The corresponding point datasets approximate surfaces where the laser beam of the scanner was reflected. Already today, there is a wide field of applications that visualize point clouds, ranging from preservation of cultural heritages to land surveying. With advances in technology, measuring techniques became increasingly precise and accurate. The associated growth of the captured point clouds offers many opportunities but also challenges. Visualizing billions of points can easily surpass the memory capacities of commercial computers. Furthermore, render times can increase, hindering the dynamic exploration and analysis. In this work, we present a processing pipeline that reduces the complexity of point clouds. As a first step, planar regions of a point cloud are approximated by rectangles. Subsequently, calculated textures and displacement maps allow a more realistic representation of the classified points through rectangles. The generated data is stored in a data structure that supports different level of detail representations based on the camera position, as well as efficient culling of invisible points. Lastly, the data structure is rendered using the OSPRay render engine. In an evaluation, different parameter configurations of our processing pipeline are examined. The collected data is analyzed in terms of render time, memory consumption, and image quality. Furthermore, all results are compared with a sphere-based reference method. Overall, a notable saving in memory consumption can be observed. Meanwhile, image quality and rendering times provide comparable results to the reference method, especially for distant to medium viewing distances. However, the achieved compression rate is dependent on the spatial properties of the point cloud. Datasets with large planar regions allow the consolidations of a high amount of points by a small number of rectangles. Meanwhile, regions with high curvature can cause overlapping geometry and a lower point reduction. All in all, the presented approach enables the rendering of point clouds at different levels of detail. Approximated geometric shapes reduce memory consumption while preserving render times and image quality. Thus, complex point clouds can be visualized more efficiently on commercial computer systems.