<p>This paper proposes an efficient Indoor Positioning System (IPS) based on a Massive Multiple Input Multiple Output (MaMIMO) communication system. The method employs a fingerprint-based approach using Channel State Information (CSI) obtained from the MaMIMO system. Efficiency is achieved through a three-stage framework combined with dimensionality reduction applied to CSI data. Singular Value Decomposition (SVD) is utilized to reduce CSI dimensionality, enabling efficient online radio map search. In the first stage, a low-resolution fingerprint map containing 4096 (4K) labeled positions and highly reduced CSI data is used to obtain a coarse position estimate. In the second stage, this estimate is refined using a small subset of the same low-resolution map with higher-dimensional CSI data than in the first stage. In the final stage, a local search is conducted around the refined estimate using a very small subset of a high-resolution fingerprint map consisting of 15,876 (16K) labeled samples. The proposed method achieves millimeter-level (8.1 mm) positioning accuracy. It is compared with state-of-the-art Convolutional Neural Network (CNN)-based approaches that require large labeled datasets and extensive training parameters. The method is also evaluated against other fingerprint-based approaches. Simulation results demonstrate that the proposed method provides superior positioning performance with significantly lower computational complexity.</p>