Improving the user peak rate in hot-spots is one of the original intention of design for 5G networks. The cell radius shall be reduced to admit less users in a single cell with the given cell peak rate, namely Hyper-Dense Networks (HDN). Therefore, the feature extraction of the node trajectories will greatly facilitate the development of optimal algorithms for radio resource management in HDN. This paper presents a data mining of the raw GPS trajectories from the urban operating vehicles in the city of Shenzhen. As the widely recognized three features of human traces, the self-similarity, hot-spots and long-tails are evaluated. Mining results show that the vehicles to serve the daily trip of human in the city always take a short travel and activate in several hot-spots, but roaming randomly. However, the vehicles to serve the goods are showing the opposite characteristics.