Due to relatively high location accuracy and reasonable simplicity, fingerprint positioning scheme is considered as a promising candidate applied in Wi-Fi based indoor environment. This paper focuses on the optimization of the positioning algorithm on the on-line stage which is critical for a fingerprint scheme. A novel approach utilizing K-Medoids and Signal Feature Extraction is proposed to reduce the computing complexity. Improved Weighted KNN algorithm which considers the physical distance is to improve the localization accuracy. Specifically but without loss of generality, we set the positioning of Wi-Fi nodes in an underground garage environment. Compared with the traditional KNN method, the experimental result shows that our scheme has relatively high algorithm efficiency and positioning precision.