Virtual light source positioning method based on indoor reflecting surface reconstruction
By generating and dynamically updating virtual light sources based on a reflective surface geometric model, the accuracy and robustness issues of visible light positioning in non-line-of-sight environments are solved, achieving high-precision indoor positioning adaptability and stability.
Patent Information
- Application Number
- CN202511650804.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-12
- Publication Date
- 2026-02-17
AI Technical Summary
Existing visible light positioning methods have poor adaptability in non-line-of-sight environments, cannot effectively utilize the geometric information of reflected signals, resulting in insufficient positioning accuracy and robustness, and are unable to adapt to dynamic environmental changes.
By establishing a geometric model of the indoor reflective surface and generating a virtual light source, the reflective surface parameters are estimated using least squares estimation and Kalman filtering. Combined with weighted positioning optimization and dynamic adaptive updating, the effective utilization of reflected signals and environmental adaptation are achieved.
It achieves high-precision indoor positioning under non-line-of-sight conditions, improves the robustness and adaptability of the system, and significantly enhances positioning accuracy and stability.
Smart Images
Figure CN121541140A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of indoor positioning technology, and in particular to a positioning method based on reflective surface geometric reconstruction and virtual light source generation, which is suitable for high-precision positioning of visible light communication (VLC) systems in complex indoor environments. Background Technology
[0002] Indoor positioning technology serves as a core support for smart homes, robot navigation, warehousing and logistics, and its accuracy and robustness directly impact the effectiveness of these applications. Among various indoor positioning technologies, visible light positioning has become a research hotspot due to its advantages such as widespread deployment of LED light sources, absence of electromagnetic interference, and high potential for positioning accuracy. The core logic of existing visible light positioning methods is to utilize the line-of-sight (LOS) path signal between the LED light source and the receiver, estimating the location using characteristic parameters such as received signal strength (RSS) and time of arrival (TOA). For example, the RSS triangulation method uses a signal strength attenuation model of different light sources to infer the distance, while the TOA positioning method calculates the path length based on the light propagation time, ultimately determining the receiver coordinates through multi-source joint solution. However, these methods all share a common drawback: extremely poor adaptability to non-line-of-sight (NLOS) environments.
[0003] In real-world indoor scenarios, walls, furniture, and people can obstruct or attenuate the line-of-sight (LOS) path. Traditional methods, relying on weak LOS signals or discarding reflected signals, suffer from a sharp drop in positioning accuracy or even fail. While some existing technologies attempt to utilize multipath signals, they have significant limitations: most simply filter reflected signals as noise, failing to delve into the underlying geometric information, resulting in low information utilization. Furthermore, existing methods generally only provide qualitative descriptions of reflective surfaces, lacking precise quantitative modeling of key physical parameters such as normal vectors and position coordinates. Consequently, it is difficult to establish stable positioning constraints, and the influence of material properties such as glass refraction and wall diffuse reflection on the light path is not considered. This leads to significant deviations between the virtual light source model and the actual situation, limiting the improvement in positioning accuracy.
[0004] Furthermore, existing methods are mostly based on static environment assumptions. When indoor reflective surfaces change due to furniture movement, temporary obstruction, or other factors, the reflection path characteristics change accordingly. However, traditional models cannot update reflective surface parameters in real time, causing positioning errors to deteriorate sharply with increasing environmental dynamism. Therefore, how to overcome the occlusion limitations of NLOS environments, accurately model the geometric relationship between reflective surfaces and reflection paths, transform invalid reflection signals into effective positioning constraints, and adapt to dynamic scene changes has become a key problem that existing indoor visible light positioning technologies urgently need to solve. This invention addresses this core need by proposing a virtual light source positioning method based on indoor reflective surface reconstruction. Through reflective surface geometric modeling, equivalent transformation of virtual light sources, and multi-source joint optimization, high-precision positioning in occluded scenarios is achieved. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention provides a virtual light source positioning method based on indoor reflective surface reconstruction, which can achieve high-precision indoor positioning under non-line-of-sight conditions and significantly improve the system's robustness, accuracy, and adaptability.
[0006] To achieve the above objectives, the present invention specifically adopts the following technical solution: This method includes the following steps: Step 1: System Initialization 1.1 Establishing the coordinate system Projection point of the center of the room to the ground Establish a right-handed Cartesian coordinate system with the origin. .
[0007] in The axis points to the ceiling. The plane is parallel to the ground.
[0008] 1.2 Hardware Deployment Install 4-6 controllable LED light sources evenly on the ceiling, with the coordinates of the light sources marked as follows: , , (This refers to the number of light sources); the receiver is equipped with a phototransistor and a high-precision clock, which can synchronously acquire the received signal strength (RSS) and time of arrival (TOA) of each optical path, denoted as... ( ),in The label “LOS” indicates the direct path, and the label “NLOS_1” indicates the first-order reflection path.
[0009] 2. Data Acquisition and Preliminary Processing For each light source, record the path characteristics of both direct light and first-order reflection, extending to second-order or higher-order reflections if necessary. Convert the RSS to an initial distance estimate: in, The LED light source emission power, For the first The received power of each path; For a 1m reference path loss, , Indoor path loss index; using TOA values to roughly screen LOS / NLOS: distance calculated from TOA. ,when When it is determined to be a LOS path, The distance is calculated based on the theoretical direct beam distance from the light source to the receiver's initial position; otherwise, it follows the NLOS path. Simultaneously, the TOA measurement noise variance is calculated. This provides a basis for measuring variance in subsequent weighted calculations.
[0010] 3. Joint estimation of indoor reflective surfaces 3.1 Plane Model Assume there are a total of main reflective surfaces indoors. The equation of a plane is uniformly written as: ,Right now in, The unit normal vector (satisfying) ), Let be the spatial coordinate vector of any point on the reflecting surface. It represents the directed distance from the reflecting surface to the origin.
[0011] 3.2 Least Squares Estimation The mirror image relationship between the starting point of the first-order reflection path and the receiving point satisfies: Construct residual vectors for all valid reflection paths: in, , The noise correction value for TOA measurement is calculated iteratively using a Kalman filter. The initial coordinates of the receiver are given; the parameters of the reflecting surface are obtained by sequential least squares iterative solution. .
[0012] 4. Virtual light source generation For each first-order reflection path, the corresponding LED light source is mirrored about the estimation plane to obtain a virtual light source: The generated virtual light source With real LED assembly Together they form an extended light source set .
[0013] 5. Weighted positioning optimization 5.1 Objective Function in, For the first Coordinates of each light source For the corresponding initial distance measurement value ; The spatial coordinate vector of the receiving end; like For a real light source (LOS path), then ; on the contrary For the corresponding virtual light source (LOS path), then ; In cases where the environment is highly dynamic or the channel fading is severe... Use a value between 0.2 and 0.5; for static scenes, Take a value of 0.5-0.8.
[0014] 5.2 Solution Method Initial coordinates of the receiver The solution was obtained by trilateration using three of the strongest LOS light sources; the objective function was solved using an improved Levenberg-Marquardt (LM) iterative method, with an initial iteration step size of 0.1m and a convergence criterion of [missing information]. (t is the iteration number), the upper limit of the iteration number is set to 20; output the final estimated coordinates. and the covariance matrix.
[0015] 6. Dynamic adaptive update 6.1 Sliding window Kalman filtering: For the coordinates of the receiving end With reflective surface parameters Perform joint state estimation; the state vector of the joint state estimation is set as follows: ,in For the coordinates of the receiving end, These are the components of the normal vector of the reflecting surface. The directed distance from the reflecting surface to the origin; process noise covariance matrix. The diagonal elements take values of The measurement noise covariance matrix R is set as a diagonal matrix based on the RSS / TOA measurement accuracy, with the diagonal elements representing the measurement variance.
[0016] Its state and observation model is as follows: State model: Observation model: 6.2 Abnormal path removal, Set residual threshold , To measure the noise standard deviation, residuals are calculated for each path. ,like If a path appears twice consecutively, it will be automatically identified as abnormal and removed; at the same time, the weights will be adjusted: the weight coefficients corresponding to abnormal paths will be adjusted. Updated to If the residual of this path is ≤ 3 times consecutively thereafter If so, the weight is restored to its original value. Attached Figure Description
[0017] Figure 1This is a geometric schematic diagram of the virtual light source construction; Figure 2 It is a system flowchart, including steps such as signal acquisition, reflector estimation, virtual light source reconstruction, and optimization solution; Figure 3 This is a positioning error distribution map; Detailed Implementation (I) System Initialization 1. Establishing a coordinate system: Establish a right-handed Cartesian coordinate system with the center of the room as the origin. The Z-axis points vertically to the ceiling (Z=0 is the ground, Z=3m is the ceiling), the X-axis is along the length of the room (X∈[0,12m]), and the Y-axis is along the width of the room (Y∈[0,10m]).
[0018] 2. Light source coordinates determination: Five LED light sources are evenly distributed on the ceiling (Z=3m), with the following coordinates: LED1: (3m, 2.5m, 3m), LED2: (9m, 2.5m, 3m), LED3: (3m, 7.5m, 3m), LED4: (9m, 7.5m, 3m), LED5: (6m, 5m, 3m) 3. Initial position acquisition of the receiver: The initial coordinates of the receiver are calculated using the three LOS light sources with the strongest signal, LED2, LED4 and LED5, and the trilateration method is used. The initial position is set as (6m, 5m, 1.2m).
[0019] (II) Data Collection and Preliminary Processing 1. Signal Acquisition: The receiver continuously acquires data for 10 observation cycles at the initial position. In each cycle, the RSS and TOA values of the direct path (labeled "LOS") and the first-order reflection path (labeled "NLOS_1", corresponding to the east wall, north wall, and ground reflection) of the 5 LED light sources are recorded synchronously.
[0020] 2. Initial Distance Estimation: Calculate the path distance based on RSS and TOA respectively, using the following formula: Distance formula based on RSS: Distance formula based on TOA: 3. LOS / NLOS discrimination Judgment threshold setting: 0.25m for medium-sized spaces.
[0021] Judgment result: When When it is determined to be a LOS path, This is the theoretical direct distance calculated based on the initial positions of the light source and the receiver; otherwise, it is the NLOS path. (III) Joint estimation of indoor reflective surfaces Reflective surface planar model settings: The planar equations of the three main indoor reflective surfaces are as follows: .
[0022] Constrained Least Squares Estimation: Residual Vector Construction: Calculate the residuals for all valid reflection paths. (iv) Virtual light source generation Based on the estimated reflective surface parameters, the five LED light sources are mirrored to generate a virtual light source, using the following formula: (v) Weighted positioning optimization 1. Construction of joint distance optimization model: Expanded light source set: 5 real LEDs + 12 virtual light sources, total number N=17.
[0023] Objective function: Weighting coefficient settings: LOS paths (5): ,in .
[0024] NLOS paths (12): ,in , .
[0025] 2. Optimized solution: Initial values: Initial coordinates of the receiver (6m, 5m, 1.2m).
[0026] Iterative method: Improved Levenberg-Marquardt (LM) iteration, initial step size = 0.1m, convergence criterion is... .
[0027] (vi) Dynamic adaptive update 1. Sliding window Kalman filtering: State vector: Covariance matrix: Process noise covariance matrix The diagonal elements take values of The diagonal elements of the measurement noise covariance matrix R are the RSS measurement variance and the TOA measurement variance.
[0028] State and observation model update: The state vector is updated once every 2 observation cycles, the receiver coordinates are updated after filtering, and the deviation of the reflector parameters is further reduced to ≤0.01m.
[0029] Abnormal path removal: Residual threshold: ,in .
[0030] Anomaly detection: If the NLOS path residual of LED3 reflected from the ground is 0.52m > 0.45m in the observation period, and the residual is 0.55m > 0.45m in the next period, it is determined to be an abnormal path, its weight coefficient is multiplied by 0.8 and it is temporarily removed.
[0031] Weighted parameters are adjusted in real time: with an indoor personnel movement speed of 0.2 m / s, the empirical attenuation coefficient of the NLOS path is adjusted from 0.4 to 0.35 to enhance the system's adaptability to dynamic environments.
Claims
1. A virtual light source positioning method based on reconstruction of indoor reflecting surfaces, characterized in that, The method comprises the following steps: (1) Establishing an indoor coordinate system, collecting spatial coordinate information of multiple LED light sources in a visible light communication system, and acquiring received signal strength (RSS) and time of arrival (TOA) characteristics from direct and reflected paths at a receiving end; wherein, the number of LED light sources is selected to be 4-6, and they are uniformly arranged on the ceiling; the receiving end is equipped with a photosensitive triode and a high-precision clock to ensure the accuracy and timeliness of signal collection; (2) According to the RSS and TOA information, the propagation distance of each light path is preliminarily estimated, and the deviation between the TOA estimated distance and the theoretical distance of the LOS path is used to distinguish the line of sight (LOS) and non-line of sight (NLOS) paths; (3) Based on the reflection geometry model of light, the normal vector and position parameters of the main reflection surface are estimated by using a constrained least squares method; (4) The known light source position is mirrored based on the reflection surface model to reconstruct the virtual light source position, which is used to represent the equivalent light source in the first-order reflection path; the virtual light source generation adopts a mirror reflection model, and the specific calculation formula is: ； wherein, is the original light source position, is the virtual light source position, is the unit normal vector of the reflecting surface, is the directed distance from the reflecting surface to the origin; when the reflecting surface is glass, mirror metal, the above mirror reflection model is directly used; when the reflecting surface is a wall, a gypsum board diffuse reflection surface, a Lambert cosine term is further introduced to model the path energy attenuation, which is reflected in the weight factor by adding an attenuation coefficient , wherein the wall surface takes 0.2-0.3, and the gypsum board surface takes 0.15-0.25; (5) A joint distance optimization model containing real and virtual light sources is constructed to estimate the spatial coordinates of the receiving end in the form of weighted least squares, and the weights are dynamically set according to the path type and channel variance; the positioning optimization model is: ； wherein, is the distance estimate from the receiving end to the light source, is the number of real or virtual light source positions, is the distance estimate from the receiving end to the light source, is the weighting factor, N is the total number of real and virtual light sources; the weighting factor is defined as: For LOS paths: where is the LOS path ranging variance; For NLOS paths: where is an empirical attenuation coefficient, is the NLOS path ranging variance; wherein, is a distance estimate, , is an original ranging value; , is an LED light source emission power, is a received power of the path, is a 1 m reference path loss, , is an indoor path loss exponent, is a TOA measurement noise correction value; (6) An adaptive dynamic updating mechanism is introduced in the positioning process, including joint state estimation of the receiving end position and the reflection surface parameters using a sliding window filter, and path anomaly detection and elimination based on residual analysis.
2. The method of claim 1, wherein, The theoretical distance is calculated based on the known light source coordinates and the initial position of the receiving end, and the initial position of the receiving end is obtained by using a three-edge positioning method for the three strongest LOS light sources; when the deviation between the TOA estimated distance and the theoretical distance is ≤ a judgment threshold, it is judged as a LOS path; otherwise, it is judged as a NLOS path.
3. The method of claim 2, wherein, The judgment threshold value is in the range of 0.1m-0.4m, and is adjusted according to the size of the indoor space: 0.1m-0.2m for small spaces (≤8m×8m), 0.2m-0.3m for medium spaces (8m×8m-15m×15m), and 0.3m-0.4m for large spaces (>15m×15m); the threshold value is set based on 3 times the TOA measurement noise variance.
4. The method of claim 1, wherein, The reflecting surface comprises a wall, a floor, a ceiling or glass, and a plane equation is wherein is a unit normal vector of the plane, satisfying ; is a spatial coordinate vector of any point on the reflecting surface; is a directed distance from the reflecting surface to the origin of the coordinate system.
1. Unit normal vector constraint: ; ②Physical position constraint: the reflection surface must pass through or be adjacent to the spatial region formed by the light source and the receiving end position; wherein, the 'adjacent' refers to the shortest perpendicular distance between the reflection surface and the'spatial region formed by the light source and the receiving end' ≤0.5m; iii. Distance non-negative constraint: all the reflection path lengths calculated based on the reflector model must be greater than their corresponding direct path lengths, and the difference between the reflection path length and the direct path length is ≥ 0.2 m; the reflector parameter estimation satisfies the following optimization model: wherein, is the position of the th light source, is the estimated distance of the th path, is the receiving end coordinate, is the total number of paths, is the unit normal vector of the reflecting surface, is the directed distance from the reflecting surface to the origin.
5. The method of claim 1, wherein, The adaptive dynamic updating mechanism specifically includes: ① sliding window Kalman filter: for receiver position and reflector parameters ; the state vector of joint state estimation is set as , where is the receiver coordinate, is the reflector normal vector component, is the directed distance from the reflector to the origin; the process noise covariance matrix is a diagonal matrix with diagonal elements , and the measurement noise covariance matrix is a diagonal matrix with diagonal elements set according to the RSS / TOA measurement accuracy; the state and observation models are as follows: State model: wherein is process noise, subject to a Gaussian distribution with mean 0 and covariance Observation model: wherein is the observation vector (containing RSS and TOA measurements), is the constructed nonlinear ranging observation function; wherein , is the received end coordinates, is the light source position, is the observation noise, following a Gaussian distribution with mean 0 and covariance R; ② Abnormal path elimination: set residual threshold ; wherein To measure the noise standard deviation, if the residual of a certain path appears twice in succession, it is automatically determined that the path is abnormal and is eliminated, and the weight is adjusted: the weight coefficient corresponding to the abnormal path is multiplied by 0.8; wherein is the actual ranging value of a single path in an observation period, is the theoretical ranging value calculated by the state vector of the observation period. ③ Real-time adjustment of path weight parameters: when the environment is dynamic or the channel is seriously fading, the weight coefficient in the path selection formula is adjusted in real time 0.2-0.5; for static scenes, 0.5-0.8 can be taken to enhance the robustness of the system; Wherein, the dynamicity is strong when the moving speed of the occluder is >0.3m / s; the static scene refers to the moving speed of the occluder <0.1m / s.