Precise Tracking of Things via Hybrid 3-D Fingerprint Database and Kernel Method Particle Filter

Precise tracking of things (PToT) using RF signals has posed a serious challenge in indoor environment. However, the need for more precision in tracking people and inventories in indoor environment has been gaining traction. Furthermore, there are demands for location-based services, such as micro-fencing for targeted advertisement. Precision is key for successful launch of the aforementioned applications. PToT relies on two main components, a novel hybrid 3-D fingerprint database and precise navigation (tracking). In this paper, we address the benefit of our novel approach in creating 3-D fingerprint database in concert with kernel method particle filter (KMPF) to achieve precise positioning and tracking. The hybrid 3-D fingerprint digital map is created from Wi-Fi and Bluetooth received signal strength as well as the fusion of optical sensor 3-D coordinate and magnetic sensor rotational attributes, hence much improved accuracy. This is the first integral component for PToT. The KMPF will track and predict the position utilizing the map information created by the high-resolution signature database. The parameter estimation for cooperative and non-cooperative modes of operations is compared with lower bound on the variance using Cramer-Rao lower bound.

I. Introduction

Precision localization for indoor mobile users pose challenges due to RF signal distortion [1] resulting from multi-path, Non-line of sight and lack of reliable GPS signal. WiFi localization is the most popular technique used in smart devices today [2], however, this technique has its serious shortcoming. It suffers from relatively large (submeter) localization error. This technique by itself, is not conducive for applications demanding high precision hence, subcentimeter localization error. Using WiFi in conjunction with other RF signaling and sensor’s data, will enable us to overcome this shortcoming, submeter accuracy. In our previous work [3], we simulated many different scenarios in order to analyze the effect of different parameters on Particle Filter performance for localization. The parameters in consideration, included number of fixed anchors, hybrid RF signaling using Ultra Wide Band (UWB) and WiFi, Moving objects sharing their location information (Cooperative, COOP) among each other versus when there is no sharing of information (Non-Cooperative, NCOOP), number of Particles, variation of Observation and State variances of the Particle filter. We are building on aforementioned work and pointing out the limitation and present novel approaches to Grid base RF signature database in conjunction with Kernel Method Particle filtering for precise navigation and localization.¹

The ranging error for UWB in our previous work [3] was modeled based on experimental data and WiFi was based on theoretical, hence deterministic.