Multi-Dimension Kernel Density Estimation for RSS Based Indoor Localization
Ching-Chun Huang, Wei-Li Huang and Hung Nguyen Manh
The illustration of the proposed learning process based on multi-dimension kernel density method.
In this paper, we proposed an improved radio signal strength (RSS) model for indoor positioning, which is also adaptive to device diversity. Typically, a RSS-based positioning system mainly relied on the concept of radio fingerprint for localization and assumed that RSS has location singularity. However, the singularity may not be available in a real application, where the RSS at a location is variant owing to signal noise, dynamic obstacles such as moving people, and device diversity. To well represent RSS distribution and consider the effects from noise, dynamic obstacles, and device diversity, in this paper, we proposed a new RSS modeling method based on multi-dimension kernel density estimation. Also, the proposed new RSS modeling method allows our system to perform region-based target localization. Comparing with the conventional point-based method, our system can utilize more information for localization. The experimental results demonstrate that our system could well model the RSS variance like signal noise, dynamic environment, and device diversity, and give better positioning performance.
The CDF of error distance for (a) the intra-device case and (b) the inter-device localization.
The maximum location error at 15 locations for the (a) inter-device and (b) intra-device cases.