DIMLOC: Enabling High-Precision Visible Light Localization Under Dimmable LEDs in Smart Buildings

论文期刊：IEEE Internet of Things Journal
论文作者：Xiangyu Liu ; Xuetao Wei ; Lei Guo

Motivation：

 With the rapid development of smart buildings, dimmable LEDs with the purpose of energy savings, which adjust their brightness based on ambient sunlight in the environment, become more popular in modern buildings.
  The motivation of this paper is to design and implement an indoor visible light localization system DIMLOC under dimmable LEDs which can cope with the blurring effects.

Problems:

• WiFi and other RF-based indoor positioning techniques have been extensively studied in prior work, which focused on methods of triangulation, received signal strength (RSS) and fingerprinting [8], [9], [10]. However, these approaches are limited in bandwidth, wireless interference, and coverage [11]. Furthermore, RF-based positioning techniques can only achieve the positioning accuracy in the meter level.
• The exist VLP techniques have different problems:
    1) The fixed frequency method only needs the fast Fourier transformation (FFT) to decode location landmarks, but it is not precise.
    2) The modulated signal method can find landmarks accurately, but it needs to design encoding and decoding schemes.

• Dimmable LEDs will have the blurring effects on images captured by the smartphone’s camera as Figure 1, which brings the challenge on determining the landmarks accurately and makes their location algorithms deteriorate.

  
  
  Fig. 1. Captured image shows the bright and dark stripes, and the blurring effects at the center.

Method:

• Proposing a novel image procssing framework that consist of efficient techniques to cope with the blurring effects caused by changes of brightness from dimmable LEDs
   1) Using a second-order polynomial fitting on each column pixel of the captured image to improve the blurring effects;
   2) Using an appropriate threshold to change the grayscale value of each pixel;
   3) Using the histogram equalization to enhance the contrast between bright and dark stripes on the captured image;
   4) Using Sobel filter to enhance edges of bright and dark stripes to decrease the impact of light noise;
   5) Using a third-order polynomial fitting for decoding landmark.
• Proposing a positioning algorithm that only requires two LEDs, which is based on vision analysis and scaling factor principles.
    1) Obtaining two angles formed by the line composed of two LEDs and the x-axis in both the world coordinate system and the camera coordinate system, respectively. We then use these two angles to position the smartphone;
   2) Compensating the displacement of image to equivalently transform the tilted image to the horizontal image for achieving the positioning.

• Choosing to use Manchester encoding. Set a 5* 80 us high level as the frame header, and a 80 us low level as the tailer. To dim the light to adapt to the environment, increasing the ratio of 0 to 75% and decrease the ratio of 1 to 25%.

  
  
  Fig. 2. Frame architecture

Experiments:

• Decoding Recognition Rate Under Nondimmable LEDs
   1) Comparing With the Modulated Signal:
   When this modulated signal is decoded without using our image processing framework (including second-order polynomial fitting, histogram equalization, and Sobel filter), we call it benchmark represented by the red line. The blue line represents the results of using our image processing architecture.
   Red line and blue line are completely overlapped.
  

  Fig. 3. Compare the decoding recognition rate
  
   2) Comparing With the Fixed Frequency:
    Transmitting different frequencies as landmarks from nondimmable LEDs. The frequencies are set to 3 kb/s, 5 kb/s, and 10 kb/s.
    In the cases of 3 kb/s and 5 kb/s, our approach has the same effect on the decoding recognition rate with the FFT method. At the case of 10 kb/s, the FFT method is slightly better than our image processing framework.
  
  Fig. 4. Comparing the decoding recognition rate

• Decoding Recognition Rate Under Dimmable LEDs
   Setting the duration time of logic 1 and logic 0 occupying 70% and 30% in each unit time, respectively. we can find that, our approach can decode landmarks correctly and the benchmark method fails.

Fig. 5. Comparing the decoding recognition rate

 Using the proposed image processing framework and FFT method to decode landmarks composed of the fixed frequency.

Fig. 6. Comparing the decoding recognition rate

• Positioning accuracy
   1) Rotated Smartphone:
   Testing the positioning accuracy under different smartphone rotation angles. The distance between the smartphone and the ceiling is fixed at 2.19 m.
  
  Fig. 7. The relationship between the rotation angle and the positioning error.
  
   2) Tilted Smartphone:
   Testing the positioning accuracy under different smartphone tilt angles. The distance between the smartphone and the ceiling is fixed at 1.69 m.The tilt angle of smartphone is set from −60 to 60 degrees.
  
  Fig. 8. The relationship between the tilt angle and the positioning error.
  
   3)Different Locations:
   Selecting randomly 40 locations in the room to test DIMLOC. Then we calculate the positioning error for each location by comparing the real positioning coordinates with the estimated positioning coordinates. Setting the distance between the ceiling and smartphone as 1.76 m.
  
  Fig. 9. The CDF of the positioning error

Conclusion:

  Compared to our previous work, this paper proposes a different image processing algorithm to cope with the blurring effects, and it also proposes an algorithm to slove the problem of tilting smartphone. However, the positioning algorithm is the same as uors which using the vision analysis algorithm.