1. Summary translation

In this paper, we propose an indoor positioning system Lightitude, the system utilizes a dense deployment of indoor light has established the existence of non-uniform intensity distribution of indoor light as a medium. Since ordinary room light due to the lack of unique features (e.g., a unique intensity or blinking frequency) can not function as a landmark, therefore, using the received light intensity (RLI) systems typically have a strong action of the user applied constraints, and the interior the ideal environment to make assumptions. Unlike these methods, we first propose a realistic model for the light intensity to reconstruct RLI distribution in a case where a given receiver any movement (position, direction), with each movement of the receiver thus collected can be used RLI positioning. Then, we designed a particle filter-based positioning module, the module's use of the natural mobility of the user to disambiguate the single RLI. The results showed that the average accuracy Lightitude in the office (720m2) and library (960m2) are, respectively 1.93m and 1.98m. Mild still be able to resist interference sunlight, shadows and a variety of user human behavior.

2. brief

The author combines a fingerprint method based on WiFi, WiFi-based understanding of the often under-AP happens. The authors town hospital, supermarket did some research and found that only a 363 m2 AP point, this is sufficient to cope with the usual connections, but not enough to use for targeting. The lighting is adequate in many indoor environments (including all fixtures, etc.), and author experiment a bit light over time, influenced by the RSSI is smaller than WiFi and CSI.

The first thought of the WiFi positioning, like fingerprints, different RLI (acceptance of light intensity) corresponds to a different position, as a classification task. However, doing so there is a major problem: The only RLI values may correspond to a plurality of possible movements and positions of the receiver (see below).

To meet these challenges, the authors used a mobile user to design the positioning module (PM). In a large indoor environment (e.g., supermarkets, schools, libraries), the mobility of the user is natural, and it provides a continuous RLI (implicitly indicating the position) and the inertial sensor data (represented implicitly direction). Using these data as input, PM generating a plurality of candidate trajectories.

In addition to module PM, author also designed a light intensity module (LIM). RLI candidates for calculating a given position and orientation (yaw, pitch and roll). The authors also considered a number of human-like shadows, unpredictable behavior and other factors

3. System

It includes data collection, LIM and PM three modules:

Data collection module

Positioning service providers to conduct a "just once" on-site investigation in the target environment. The purpose of this survey is to obtain a light intensity characteristic of indoor lighting. To achieve this goal, the service provider will receive the device (eg smartphone) placed near each lamp for a few seconds. LIM coordinates of the recording light, and the coordinates and the direction vector of the receiver RLI.
If the user opens the "brightness" application, it is equipped with a receiver to collect the gyro signal from the optical sensor, and the gyro, accelerometer and direction sensor inertial sensor data collected. These data are Lightitude input. IMU data sampling rate of 67Hz, and the light sensor changes to follow the sensor rules.

Light intensity model (LIM):

The type of input data, LIM has two tasks:
• LIM service provider using the data as input, wherein the light intensity is calculated for all indoor illumination lamp.
• LIM using the data from the user and lighting features (coordinates, strength characteristics) as input, calculates RLI candidate positions.

Positioning module (PM):

PM generating a set of candidate objects (i.e., recommended particle filter particles) in indoor environments, and the theoretical RLI (from LIM) and the ground-truth RLI (from the light sensor). If the theory RLI particles close to the ground-truth RLI, it will gain more weight. By using the user's mobility, the right to update the particle weight, until a trigger convergence rules.

4. The light intensity model (LIM):

OF in the original LIM made some improvements, which the main improvement is to consider a plurality of light receiving points and a terminal capable of receiving the ratio c. See the original concrete.

The positioning module (PM)

该部分数学公式太多，读者可参考原文，大致意思是PM在室内环境中生成一组候选对象（即，建议的粒子过滤器中的粒子），并将其理论RLI（来自LIM）与ground-truth RLI（来自光传感器）。如果粒子的理论RLI接近ground-truth RLI，则它将获得更大的权重。通过利用用户的移动性，更新粒子的权重，直到触发收敛规则。并且考虑了阳光的干扰，人体阴影，不可预测的活动。