A Dynamic Hybrid Projection Approach for Improved Wi-Fi Location Fingerprinting

Projection techniques have been used in Wi-Fi location fingerprinting systems to improve positioning accuracy. However, environmental dynamics present challenges to projection design. Furthermore, current projection-optimization techniques used in positioning, such as principal component analysis (PCA) and multiple discriminant analysis (MDA), have both advantages and limitations. This paper proposes a dynamic hybrid projection (DHP) technique for improved Wi-Fi localization, in which the projection is dynamically determined by simultaneously exploiting the complementary advantages of PCA and MDA while avoiding their unfavorable properties. The main contribution of this work is twofold: First, this study provides a novel formulation of a hybrid projection, which embeds the discriminative power into PCA and compensates for the two numerical problems of MDA in a unified framework. Second, DHP dynamically adjusts the hybrid mechanism with additional information, regarding the online-input region. That is, the proposed projection is input dependent, whereas traditional projections are fixed after training. This study applies the proposed algorithm to location fingerprinting in a realistic indoor Wi-Fi environment. On-site experimental results demonstrate that DHP outperforms static projection schemes, reducing the 50th and 67th percentile localization errors by 24.73%-30% and 18.18%-19.51%, respectively, compared with PCA and MDA.