Advanced visible light-based indoor positioning techniques : development and analysis
Date of Issue2019-06-06
Interdisciplinary Graduate School (IGS)
Energy Research Institute @NTU
In recent years, light-emitting diodes (LEDs) enabled indoor visible light positioning (VLP) technology has attracted ever-increasing attention, due to its numerous potential applications including pedestrian localization and navigation, robots monitoring, assets tracking and even industrial manufacturing. Compared with traditional radio-frequency (RF)-based systems (e.g., Wi-Fi, Bluetooth), LED-based VLP systems has many inherent advantages, such as energy efficiency, long life-time, better positioning accuracy, etc. Besides, it can be integrated with visible light communication (VLC) system which is a fast-developing next-generation wireless communication system. Nevertheless, the development of accurate, stable and low-cost indoor VLP systems faces many challenging issues: i) limited mobility and robustness due to the line-of-sight constraint; ii) conflict of the complex processing and positioning algorithms with the limited memory, power and calculation ability of mobile devices; iii) high positioning accuracy to centimeter, even to millimeter level; iv) interruption/outage during the localization; v) the integration of VLC and VLP systems. This thesis addresses part of the pressing issues mentioned above to improve the overall performance of a VLP system. Both photodiodes (PDs) and camera/image sensors (ISs) are commonly used as the receivers of a VLP system. In an IS-based VLP system, a target’s location is calculated by solving a camera projection model which describes the one-one correlation between an object point and its projected image point. However, the conventional iterative solvers, e.g., Levenberg-Marquardt (LM), which are used to solve the projection model, are found sensitive to the initial guesses. It means the algorithm may fail to converge or suffer a long response time and a large positioning error when starting from a bad guess. To speed up the positioning process and guarantee the success rate, we derive a closed-form solution to the receiver’s position and orientation using singular value decomposition (SVD) technique and propose an SVD based positioning algorithm. Extensive tests verify that the proposed algorithm is 50-80 times faster than the conventional LM algorithm and avoids convergence failures caused by the bad initial guesses. Meanwhile, we derive the Cramer-Rao lower bound (CRLB) as the theoretical accuracy limit. Simulations are performed to reveal the impact of the various system parameters on the positioning error bound. A robust and fast 3D IS-based VLP system with centimeter-to-decimeter level accuracy is demonstrated experimentally. On the basis of the above IS-based VLP system, a sensor fusion-based positioning algorithm is proposed to enhance positioning accuracy by effectively fusing data from the image sensor and the motion sensors, both of which are widely accessible in common mobile devices. The issue of sensor fusion is formulated as a multi-objective non-convex optimization problem and mathematically solved using SVD technique, again. Experiment results show that the proposed singular value decomposition-based sensor fusion (SVD-SF) algorithm brings approximate 44% positioning accuracy improvement compared with the algorithm employing a single image sensor under the same experimental settings. Simulations are conducted to further evaluate its positioning error performance under different noises conditions. The IS-based VLP systems mentioned above requires capturing at least three LEDs in every picture to calculate the targets’ positions, otherwise positioning is interrupted or fail. The requirement greatly reduces the system robustness and flexibility. To deal with this problem, we propose a single circular LED positioning (SCLP) system where a common circular LED lamp with a point marker is used as a transmitter. The LED image is no longer treated as a point as in the existing works, but as an image whose geometric features are exploited to determine the receiver’s orientation and location relative to the reference LED lamp. The expressions for determining the receiver’s orientation and location are derived in terms of the geometric parameters. Our algorithm effectively enhances the system robustness by reducing the minimum number of required LEDs. The SCLP system is experimentally validated and centimeter-level positioning accuracy is achieved in an area of 3m × 3m. The IS-based VLP system can be classified as a self-localization system where targets detect, process and calculate the positions of themselves. This system structure may not be appropriate for portable mobile devices having limited power, memory and computation capability. Besides, the self-localization system lacks the overview of a global view of real-time locations of all targets on the map which is required in some application such as simultaneous navigation and monitoring of multiple targets. Considering these problems, a multi-target localization system is proposed for the first time in terms of visible light positioning (VLP). PDs are mounted on the ceiling to detect light signals from mobile targets (MTs) that carry LEDs, while a central server processes the received signals and calculates the positions of MTs and sends their location information back to them. By exploiting the sparse nature of a localization problem, the multi-target localization is converted into a problem of sparse matrix reconstruction. A 3-step workflow is developed to solve the problem by employing the compressive sensing (CS) theory. The proposed system and supporting algorithms are verified through extensive simulations. The impact of various parameters including the PD number, SNR, grid square size on system positioning performance is investigated.
DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Signal processing