Place recognition, which allows distinguishing one location from another, is an extremely challenging problem. Current approaches have limitations such as susceptibility to large environmental changes, learning requirements, and recognition latency. This paper presents a cloud-based place recognition system that does not require any form of learning. The proposed method refines the cloud database by segmentation of 3D maps to a set of sub-maps and distinguishes them according to their highest z-coordinate points as an efficient search algorithm. In addition, with the presented technique, implementation depends on a parallel computation architecture to speed up the complex stages of the place recognition. In order to evaluate the presented approach, the experiments are carried out with publicly available 3D scan datasets. The results are compared to state-of-the-art techniques, which prove that the computational cost is greatly decreased regardless of the size of the compared maps.