High-Definition (HD) maps have become an important research topic during the past decade because they are one of the main components of automated driving systems. Point clouds are considered the primary data source for creating and updating those maps. However, road extraction from these data is still challenging due to the different attributes of point clouds and the diversity of road types and environments. Deep Neural Networks (DNN) have recently become popular in road extraction from point cloud data. Nevertheless, the model training requires high computational capacities and lots of labeled data. This paper proposes a method that involves using Shallow Neural Networks (SNN) for road pavement point extractions from 3D colored point cloud data. This method is applied directly to the raw point clouds providing point-wised labeling. The RGB values of the point cloud were used as input. The method performance was assessed for a rural road sector with a length of more than 1 km. The road point cloud was generated based on UAV drone images. The road points were extracted at average completeness, correctness, quality, and overall accuracy of 98.36%, 99.53%, 97.90%, and 99.87%, respectively. The approach shows notable improvement when compared to various state-of-the-art methods.