Skip to main content
Open Access Publications from the University of California

UC Riverside

UC Riverside Electronic Theses and Dissertations bannerUC Riverside

Indoor Positioning System Using Visible Light Communication and Smartphone With Rolling Shutter Camera


Indoor positioning systems provide location based service within buildings. Because the Global Positioning System is usually unavailable in indoor environment, other positioning technologies making use of optical, radio or even acoustic techniques are used for indoor positioning application. In this dissertation, an indoor visible light communication positioning system using a smartphone with rolling shutter camera is proposed. The LED transmits periodical signals with different frequencies high enough to avoid flickering as its optical tags. The camera exploits the rolling shutter effect to detect the fundamental frequency of optical signals. This kind of systems use smartphone camera as the receiver with no requirement of extra hardware. And at the same time, the optical communication link allows a date rate much higher than the frame rates of traditional optical camera communication. For the work of this research topic carried out so far, the roles of camera parameters determining rolling effect performance are studied and a technique to measure the camera readout time per column is presented. Factors limiting the detectable frequency range is explained based on the discussion of rolling shutter mechanism. Followed is the analysis of frequency detection reliability and resolution with Fourier spectrum analysis. After that, we present a background removal algorithm to recover the modulated information from background interference. In our algorithm, only one extra image is used and a one-time image alignment is required. Linear filter performs poorly in this case because the image with rolling shutter effect is a multiplication of modulated signal and the reflectance of the scene. Therefore we use background division rather than background subtraction to remove the background interference. The modulated illumination pattern is analyzed and the performance of background removal is evaluated by experiments. The experiment result shows that the interference is significantly compressed after background removal.

After the LED ID being detected successfully, an indoor positioning algorithm is proposed using multi-view imaging geometry with built-in smartphone inertial sensors. Due to the narrow field of view (FOV) of camera and the illumination placement of buildings, there is usually only one LED can be captured within camera FOV at the same time. In our proposed algorithm, only one detected LED with known position and one camera are required. And this system can still work when the VLC link is temporarily blocked or the LED temporarily moves out of the camera FOV. Another advantage of proposed algorithm is that no additional accessory will be needed except those sensors built inside the smartphone. In addition, the position is calculated at the receiver end without the knowledge of transmitter specifications, therefore there is no privacy concerns. The low-cost built in IMU sensor exhibits significant systematic errors, axes misalignment and noisy measurement. Even though a IMU calibration process can eliminate the systematic errors and axes misalignment, the noise remains even after calibration and could magnificently degrade the system performance. To maximize a posterior of obtained measurements, Kalman filter is applied for the sensor fusion by combining the system kinematic prediction and new measurement. Due to the nonlinearity of this system, an extended Kalman filter is used to correct the system model. A simulation is conducted to verify the proposed system. The simulation data of accelerometer and gyroscope are generated based on real world measurement from accelerometer and gyroscope. The simulation results demonstrate that the position error and orientation error are well bounded in the 3 sigma bounds and the maximum position error observed during the 2 minutes simulation is 0.1941 m over 50 averaged Monte Carlo trials.

Main Content
For improved accessibility of PDF content, download the file to your device.
Current View