A Survey of Recent Development in Indoor Visible Light Positioning

介绍

本文是作者就读于深圳大学计算机与软件学院期间，修读伍楷舜教授讲授的《专业研究英语》时所写的期末论文作业。研究内容为可见光定位技术在近年来的发展。

Abstract

Due to the limitations of outdoor positioning systems (such as GPS) in the indoor environment, indoor positioning technology has sprung up. In recent decades, academia and industry have proposed many feasible indoor positioning solutions. Among them, Visible Light Positioning (VLP) technology shines because of its low cost, high bandwidth, low power consumption, and high longevity characteristics. This paper introduces the latest frontier VLP systems, summarizes and classifies them according to the specific technology they use, and focuses on Pulsar. Although there are still unresolved problems in this field, we still firmly believe that in the future VSP will become an important, feasible, low-cost, low-energy, and efficient solution that matches urban intelligence.

Background

When it comes to positioning technology, people often think of GPS navigation and so on. It is believed that outdoor positioning technology is relatively mature, such as GPS, China’s Beidou positioning system, and cellular wireless network positioning system, etc., which can provide users with outdoor positioning services with sub-meter accuracy. However, with the continuous advancement of urbanization, cities are becoming smarter where complex buildings such as large indoor public places, office buildings, residences, and apartments, including large commercial centers and underground parking lots, are increasing day by day. As a consequence, people’s demand for positioning technology is no longer satisfied with outdoor positioning, but gradually extends to indoor spaces.
However, GPS struggles to operate reliably indoors due to signal degradation and multipath propagation [1]. Outdoor positioning technology cannot be directly applied to indoor positioning. Therefore, indoor positioning technology came into being.
In terms of demand, indoor positioning requires higher accuracy and faces the test of more complex environmental factors. The outdoor channel environment is relatively simple, while indoor building structures and building materials will cause complex multipath effects. Wireless signal propagation is greatly affected by non-line-of-sight, and mobile terminals lack hardware support. All these have caused great difficulties for indoor positioning. After two decades of exploration in academia and industry, a variety of relatively mature indoor positioning technologies have been formed.
Among them, Visible Light Positioning (VLP) technology has many advantages such as low cost, high bandwidth, low power consumption, high longevity, etc. [2]. But at the same time, like all other indoor positioning technologies, VLP has inevitably faced many challenges, which are listed in Table 1[2].
In order to solve these challenges and achieve efficient visible light positioning, the next section will introduce the related work carried out by academia and industry in recent years.

Introduction of Related Works

According to a recent survey, there are many solutions for indoor visible light positioning. In each solution, at least there are light sources and sensors, which will be the main components of the visible light positioning system and an important basis for the classification of related tasks at the hardware level. On the software, various related research work on visible light positioning systems, different light source feature extraction, optical signal processing methods, database construction and feature matching methods have been tried in different ways. In these attempts, the system framework and construction process are also different [2]. In this section, we will classify and review these works.

Light source
In related works, light sources are roughly divided into Modified Light Source and Unmodified Light Source. Among the two types, the former requires special strict design, while the latter has more relaxed requirements for the light source. Different solutions also have different requirements for the shape of the light source (Fig. 1). In order to reduce the error, some solutions require that the point light source must be designed separately; some can use any light source [3] after being processed by a specific Point-source Approximation algorithm.
Receiver
The type of receiving device can also be another critical design factor in VLP systems [4]. Generally, there are two types of receivers in VLP systems: photo-detector (or called photo sensor, such as photo-diodes) and imaging sensor (or camera sensor, such as smartphone camera). So we can also classify VLP systems into photo-detector-based and imaging-sensor-based. LiTell uses a camera to capture light information [5], and Pulsar is an improved version of LiTell, which is based on many principles Inspired by LiTell, but used PD based sensor to receive light signal [3].
Light signal characteristics
In order to distinguish the direction and attributes of the light emitted by the base station, it is necessary to extract the information in the light. For a scene, if you can make full use of the light information, you can determine the relative direction and distance between the recipient and the light source. Therefore, information that can identify and distinguish multiple beams of light can locate a position more accurately.
In order to determine the geometric direction of the light, the existing hardware can extract the AoA (angle-of-arrival) of the receiver and the light source to determine the direction of incidence (Fig. 2); in order to determine the distance factor between the light source and the receiver, it is necessary to extract RSS (Received Signal Strength), etc.; to determine the intrinsic properties of the light source, it needs to be extracted CF (Characteristic Frequency) of the light source [3, 5].
Database establishment and signal feature matching method
At this level, the existing visible light positioning solutions can be roughly divided into two categories, fingerprint identification methods and model-driven methods.
The fingerprint identification method associates each positioning point with a set of characteristic parameters provided by radio frequency signals of different base stations, and builds a fingerprint database based on this, and realizes positioning by matching the signal received by the target to be measured with the database. Real-world tests showed that such features lack stability in the presence of human activities and environmental changes, and are often not discriminative enough to offer fine location granularity [6].
A model-driven method that calculates the distance between user equipment and infrastructure landmarks through received signal strength (RSS), phase, or propagation time models. Both Pulsar and LiTell belong to the visible light positioning achieved by the typical model-driven method.
The main works related to the VSP system in recent years is listed in Table 2. [2].

Key Studies

In this survey, I have a strong interest in Pulsar, a PD-Based AoA Sensing VLP system. Because Pulsar largely solved the shortcomings existed in recent decades the VLP system: for example, require specialized LED, which prevents large-scale deployment; for example, need a camera, a high-power short-covering, leading to unsustainable positioning. As a result, I take Pulsar as my key research object.
Pulsar uses PD as its light sensor, which reduces its power consumption by several times compared with a camera-based VLP system [3]. In addition, due to the much higher dynamic range of the PD, Pulsar can capture the natural frequency of the LED, which is much weaker than that of the FL, even when the LED distance exceeds 9m, which is in sharp contrast to the camera-based solution that can only work in one range. Even for FL, the range is 2.5m. This remarkable feature allows Pulsar to capture multiple lights at the same time, even if these lights are sparsely deployed on high ceilings and are positioned without any blind spots. These benefits also bring new challenges: Near-far interference and Lack of spatial resolution.

Pulsar’s key solution principle lies in a novel mechanism, referred to as sparse photogrammetry [3]. Sparse photogrammetry derives the Angle of Arrival (AoA) of the light sources based on the compact light sensor. The sensor mainly contains two photo-diodes with different Fields of View (FoV). The differential response between them follows a non-linear function with the AoA. This can be calibrated and known beforehand when it is manufactured. By measuring these responses, it can map them to the light source’s AoA. Pulsar uses AoA in place of RSS to find its way out of the Lambertian model. This enables Pulsar to localize lights of any shape. Using a triangulation model, it can find out the device’s 3D location. If there are more than three lights, the Pulsar sensor can also find out the orientation angles [2].
The deployment and maintenance of Pulsar is very simple. Pulsar only requires a onetime bootstrapping procedure, where a surveyor walks across the lights inside a building, while the lights’ frequency features are captured by Pulsar, and lights’ locations automatically registered on a publicly available floor map and stored in a database. Following any subsequent change of light fixtures, the database can be incrementally updated based on user devices’ feedback.

Discussion

It can be seen from the above that in previous studies, many visible light positioning systems have been given. Researchers of these systems have used genius designs and proposed a variety of visible light positioning solutions. How to evaluate their work? It should be evaluated by indoor positioning performance indicators [7]:

Positioning accuracy refers to how close the estimated position of the space coordinate is to the real position;
Penetration refers to the property of one substance passing through another substance;
Power consumption, power loss, refers to the difference between input power and output power of equipment, devices, etc.;
Anti-interference, interference refers to the system or technology that causes damage to the reception of useful signals. Anti-interference is a system or technology used to counter any interference in communications or radar operations;
Deployment cost refers to the resources consumed during the installation and deployment of the system, including expansion time, space, capital, etc., and is a standard for measuring related indicators;
Transmission distance refers to the farthest transmission distance of the equipment under the control of the distributor. If the distance is exceeded, the signal will be lost.

First, compare the two positioning modes. Fingerprinting based approach is arguably the most accurate way to realize radio-based localization. Using RSS or more detailed physical-layer information, the location granularity can reach 1ft2. However, suffering from environment changes, human presence/mobility, etc., RF fingerprinting is highly unreliable typical 80-percentile error fall in 6 to 8 meters. More importantly, the training overhead is unbounded because under multipath effects, nearby locations do not have a deterministic signature correlation. For example, latest WiFi fingerprinting method needs to employ a robot to survey each 1 ft2 location spots for 100 times to average out signature variations. The fingerprints will soon become stale in practical environments [5].

Model-driven can avoid labor-intensive fingerprint recognition, and there is no special requirement for the design of the luminaire. It can be quickly deployed in an arbitrary indoor space, such as Pulsar and Litell, which implement this mechanism. Therefore, the deployment cost caused by the model-driven approach is very low [3, 5]. On the other hand, the update and maintenance cost of the model-driven method is very low. Even if the information of the light source is gradually changing, it can also update the database in real time according to the light source information obtained by the user during use. For example, the pulsar team realized automatically registering light landmarks [3]. But due to inherent vulnerability of RF signals, the robustness problem remains unsolved [8].

The receivers that capture the light source signal mainly include cameras and PD based sensors. The use of cameras has a good advantage, because all smartphones now have a camera for selfies on the side of the screen. This is beneficial to the large-scale promotion of visible light positioning technology, because mobile terminals have ready-made hardware foundations. However, the power consumption of the camera for a long time is too large, so the pulsar team has to introduce the PD based sensor for improvement. If this solution is widely accepted, smart phones in the future may additionally design a PD based sensor on a certain corner of the front of the screen [3].

In addition, I think the actual effect of the existing solutions and the correctness of the conclusions are open to question. On the one hand, whether it is a fingerprint recognition method or a model-driven method, there is more or less the problem of lack of data sets [9]. Furthermore, most of the works do not discuss the ground truth recording methodology when, in most cases, reporting mean localization errors of 50 mm or lower. This also casts doubt on the accuracy of such reported measurements [9].

Open Issue

There has been a lot of progress in the field of VLL systems in recent years, but there still remains open issues that need to be dealt with.

Line of Sight (LoS) Problem. One of the major concerns with all the systems is the line of sight problem. Anything blocking the line of sight between the transmitter and the receiver is halting the whole system or significantly affecting the accuracy. In EyeLight [10], the LoS problem is addressed by leveraging shadows, but the accuracy is not as good as those with LoS. To find a way to solve the LoS problem with better accuracy is still a challenge.
Integration with Other Sensing/Localization Techniques. Building new localization systems fusing multiple techniques along with visible light to gain more accuracy is also a promising path for future research. Note that [11, 12] have used IMU sensing data to enhance their performances. Wang et al. [13] have exploited the bi-modal magnetic field and ambient light data obtained by smartphone magnetic and light sensors for indoor localization with a deep learning approach based on LSTM (long short-term memory).

Conclusion

In this article, we briefly introduce the background and development, connotation and extension of visible light positioning technology, and classify and review some important frontier related work in recent years through the differences in light sources, receivers and other elements. In particular, we conducted a key research on Pulsar. Then, we discussed and thought about VLP technology. From the current point of view, although VLP is becoming more and more mature, it may not become the only indoor positioning solution in the future. In the future, VLP is more likely to shine in the integration with other technologies.

Reference

Gu Y, Lo A, Niemegeers I. A survey of indoor positioning systems for wireless personal networks[J]. IEEE Communications surveys & tutorials, 2009, 11(1): 13-32.
Rahman A B M, Li T, Wang Y. Recent advances in indoor localization via visible lights: A survey[J]. Sensors, 2020, 20(5): 1382.
Zhang C, Zhang X. Pulsar: Towards ubiquitous visible light localization[C]//Proceedings of the 23rd Annual International Conference on Mobile Computing and Networking. 2017: 208-221.
Pathak P H, Feng X, Hu P, et al. Visible light communication, networking, and sensing: A survey, potential and challenges[J]. IEEE communications surveys & tutorials, 2015, 17(4): 2047-2077.
Zhang C, Zhang X. LiTell: Robust indoor localization using unmodified light fixtures[C]//Proceedings of the 22nd Annual International Conference on Mobile Computing and Networking. 2016: 230-242.
Liu H, Gan Y, Yang J, et al. Push the limit of WiFi based localization for smartphones[C]//Proceedings of the 18th annual international conference on Mobile computing and networking. 2012: 305-316.
RAO Wenli, Overview of Indoor 3D Positioning Classification, Methods and Techniques.
Lymberopoulos D, Liu J, Yang X, et al. A realistic evaluation and comparison of indoor location technologies: Experiences and lessons learned[C]//Proceedings of the 14th international conference on information processing in sensor networks. 2015: 178-189.
Glass T, Alam F, Legg M, et al. Autonomous Fingerprinting and Large Experimental Data Set for Visible Light Positioning[J]. Sensors, 2021, 21(9): 3256.
Nguyen V, Ibrahim M, Rupavatharam S, et al. Eyelight: Light-and-shadow-based occupancy estimation and room activity recognition[C]//IEEE INFOCOM 2018-IEEE Conference on Computer Communications. IEEE, 2018: 351-359.
Li L, Hu P, Peng C, et al. Epsilon: A visible light based positioning system[C]//11th {USENIX} Symposium on Networked Systems Design and Implementation ({NSDI} 14). 2014: 331-343.
Zhao Z, Wang J, Zhao X, et al. NaviLight: Indoor localization and navigation under arbitrary lights[C]//IEEE INFOCOM 2017-IEEE Conference on Computer Communications. IEEE, 2017: 1-9.
Wang X, Yu Z, Mao S. DeepML: Deep LSTM for indoor localization with smartphone magnetic and light sensors[C]//2018 IEEE International Conference on Communications (ICC). IEEE, 2018: 1-6.