CN114268918B - Indoor CSI fingerprint positioning method for rapid off-line library building - Google Patents
Indoor CSI fingerprint positioning method for rapid off-line library building Download PDFInfo
- Publication number
- CN114268918B CN114268918B CN202111342171.9A CN202111342171A CN114268918B CN 114268918 B CN114268918 B CN 114268918B CN 202111342171 A CN202111342171 A CN 202111342171A CN 114268918 B CN114268918 B CN 114268918B
- Authority
- CN
- China
- Prior art keywords
- csi
- fingerprint
- moving carrier
- target detection
- indoor
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Abstract
The invention provides an indoor CSI fingerprint positioning method for fast off-line library establishment, and belongs to the technical field of indoor positioning. The method comprises the following steps: the method comprises the steps that a moving carrier is used for collecting CSI and position information of all places of an indoor scene in an off-line mode, and video images of the moving carrier are shot; the training target detection model is used for detecting a moving carrier in a video image; analyzing the position information of the moving carrier by using a trained target detection model, combining target tracking with target detection, and introducing a weighted prior frame for target detection by using a target according to a result; and carrying out time synchronization on the collected CSI and the position of the motion carrier, carrying out smoothing processing and sampling on a CSI sequence, constructing a CSI fingerprint and matching with the position coordinate, constructing an offline fingerprint map library, and carrying out online detection based on the fingerprint library. The method reduces the complexity of off-line library construction, and has low cost and high precision for fingerprint library construction.
Description
Technical Field
The invention belongs to the technical field of indoor positioning, and particularly relates to a method for positioning indoor Channel State Information (CSI) fingerprints, which realizes rapid off-line library establishment based on a computer vision technology.
Background
In the indoor positioning field, fingerprint positioning becomes a current research hotspot because of the potential of high-precision positioning. Fingerprint positioning is divided into two links of off-line library establishment and on-line positioning: in the off-line library building step, signal fingerprints at all reference points need to be collected, and an off-line fingerprint map library is built. In the online positioning link, signal fingerprints acquired by the points to be measured need to be matched with fingerprint vectors in a fingerprint map library, reference points with high fingerprint similarity are found, and the positions of the points to be measured are estimated by using position coordinates of the reference points.
The common scheme of the offline library building link is manual library building, and the scheme needs to grid an area to be positioned, select a representative point (reference point) in each grid area, measure the signal characteristics of the representative point, and generate a signal fingerprint after processing. Finally, the positions of the reference points are matched with the corresponding signal fingerprints one by one and stored, and the fingerprint map library can be obtained. The huge overhead problem of the process also becomes the bottleneck of the wide application of fingerprint positioning.
Therefore, some schemes for automatic library creation are proposed. For example, a crowdsourced fingerprint acquisition scheme uses a mobile device sensor carried by a user to track its position and simultaneously acquire a signal fingerprint, thereby constructing a fingerprint map library (reference [1 ]). Although the library building method based on crowdsourcing alleviates the problem of library building overhead to a certain extent, the process still has difficulty in avoiding participation of a large number of people, and the process is susceptible to environment and people, so that a high-precision fingerprint library is difficult to obtain.
Another automated banking scheme relies on regularly collecting signal fingerprints by sensors or agents distributed indoors, which avoids human involvement (reference [2 ]). However, both the delivery of a large number of smart devices and the acquisition of the location of the smart devices themselves result in a large financial investment and a high system complexity for the method. Therefore, the research on the automatic fingerprint database establishment scheme with low complexity and low cost is an important task for determining whether the fingerprint positioning system can be widely applied in practice.
Reference:
[1]Zhao W,Han S,Hu R Q,et al.Crowdsourcing and Multisource Fusion-Based Fingerprint Sensing in Smartphone Localization[J].IEEE Sensors Journal.2018,18(8):3236-3247.
[2]Gucciardo M,Tinnirello I,Dell'Aera G M,et al.A Flexible 4G/5G Control Platform for Fingerprint-based Indoor Localization[C]//IEEE INFOCOM 2019-IEEE Conference on Computer Communications Workshops(INFOCOM WKSHPS).IEEE,2019.
disclosure of Invention
The invention provides an indoor CSI fingerprint positioning method for fast off-line database building, aiming at solving the problem of huge expense in the link of off-line database building for fingerprint positioning, so that computer vision-assisted automatic database building is realized, a large amount of manpower or material resource investment in the link of fingerprint database building is relieved, and the usability of a fingerprint positioning system is improved.
The invention discloses an indoor CSI fingerprint positioning method for fast off-line library establishment, which comprises the following 4 steps:
step 1, collecting indoor scene CSI and position information in an off-line manner;
and fixing the transmitter and the camera in a scene, making the moving carrier carrying the receiver move around in an indoor environment, and acquiring the CSI of each reference position and the video image of the moving carrier.
Step 2, collecting and constructing a training sample, and training a target detection model off line;
acquiring a video image according to the mode in the step 1, wherein a training sample is obtained by sampling a video frame and calibrating the position of a motion carrier in the image; and the target detection model adopts a fast RCNN model, video images are input into the model, and the position of the output motion carrier is detected.
Step 3, analyzing the position information of the moving carrier;
and carrying out target detection on the video image by using the trained target detection model at fixed time intervals, and tracking the target in the time interval.
Introducing a weighted prior frame to a target detection model, specifically: after target tracking is finished at each time interval, a tracking result is utilized to provide position prior for target detection, and pixel points arranged near a target area last time have higher prior weight.
And 4, constructing an off-line fingerprint map library, and carrying out on-line positioning based on the fingerprint map.
In the step 4, the step of constructing the off-line fingerprint map library comprises the following steps:
(4.1) carrying out time synchronization on the collected CSI and the position of the moving carrier;
(4.2) smoothing the CSI sequence;
(4.3) sampling the CSI sequence;
the time sampling is performed based on the CSI similarity, and specifically includes: setting a CSI similarity threshold range, setting the last sampling point as A, taking the A as a current point, finding the next candidate sampling point B through step length step by the A, and calculating the CSI similarity between the A and the B; if the similarity between the A and the B is higher than the upper limit of the threshold value, discarding the B, taking the B as a new current point, and continuously searching a next candidate sampling point by step length; if the similarity between the A and the B is lower than the lower threshold, setting step size to be halved, and searching the next candidate sampling point by taking the A as the current point again; if the similarity of A and B is within the threshold range, taking B as the next legal sampling point of A;
and (4.4) extracting amplitude and phase information of the CSI from each CSI sampling point to construct a CSI fingerprint, and matching and storing the CSI fingerprint with the position coordinates at corresponding moments one by one to obtain an offline fingerprint map library.
Compared with the prior art, the invention has the advantages that:
in the face of the problem of huge expense of fingerprint library construction, the invention utilizes the target detection and target tracking algorithm which is developed relatively well nowadays to shoot a moving carrier carrying a receiver by using a camera in a fingerprint positioning off-line library construction link, and then analyzes video information to automatically obtain the target position, thereby reducing the complexity of off-line library construction. Meanwhile, the invention utilizes the spatial correlation of the target in a short time, combines the target tracking algorithm and the target detection algorithm and constructs an automatic fingerprint database establishing system with low cost and high robustness.
Drawings
FIG. 1 is a general flow chart of an indoor CSI fingerprint positioning method for fast off-line library building according to the present invention;
FIG. 2 is a schematic diagram of an experimental scenario of the method of the present invention.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and examples.
The invention provides an indoor CSI fingerprint positioning method for fast off-line database building, which is used for realizing automatic off-line database building based on computer vision assistance and building an indoor CSI fingerprint positioning system based on the automatic off-line database building. As shown in fig. 1, the method of the present invention comprises the following five steps.
First, collecting CSI and position information off line.
The invention is based on a wireless communication system, for example, a receiving and transmitting device of Wi-Fi, LTE,5G and the like collects signal CSI.
The experimental scenario set by the embodiment of the present invention is shown in fig. 2: the transmitter Tx is fixed in position, so that a moving carrier carrying the receiver, such as a trolley, can move around in an indoor environment, CSI at each position is acquired, and meanwhile, a high frame rate camera records corresponding picture information. In a later process, the frames can be used to acquire the position information of the trolley for constructing a fingerprint library. In the embodiment of the invention, the camera is positioned at the high position of the lower right position of the experimental scene, and can shoot the whole experimental area in a overlooking angle.
During a specific experiment, the following recommendations were provided:
(1) The camera is fixed and adopts a top view angle as far as possible to shoot the motion carrier so as to fully utilize picture information.
(2) For the movement mode of the trolley in the area to be positioned, because random walk is difficult to well cover a two-dimensional plane area, undersampling or oversampling of a reference point is easily caused in some areas, and therefore a regular movement track is more recommended to be adopted for collecting the reference point, such as a grid track which is staggered transversely and longitudinally.
(3) Since the high resolution of the CSI to the location mainly comes from the unique multipath reflection characteristics of each location, the direct path in the LOS (Line of Sight) environment can greatly interfere with the CSI features with spatial specificity brought by each reflection path, so it is recommended to use the NLOS (Non Line of Sight) scene for CSI fingerprint positioning. As shown in fig. 2, a metal baffle can be used to block the direct path of the transmitting antenna to the experimental area to create NLOS conditions.
(4) As a moving carrier travels indoors, the antenna carried by the carrier may be in multiple orientations. Therefore, the moving carrier is required to have a high circular symmetry to prevent it from affecting the surrounding environment of the transmitting antenna when it is in different orientations, and the transmitting and receiving antenna should have a good enough omni-directionality.
And secondly, collecting and constructing training samples, and training a target detection algorithm model in an off-line manner.
Firstly, constructing a training sample, acquiring a wandering video of a motion carrier in multiple scenes in advance, sampling video frames to obtain multiple pictures, and calibrating the position of the motion carrier in the pictures to be used as the training sample of a target detection model.
The training target then detects the neural network and saves the model.
The purpose of this step is to train a target detection model for automatically resolving the position of a motion carrier from a video in a subsequent link, so that developed and mature target detection algorithms including fast RCNN (regions with a connected neural network), YOLO (young Only Look one), SSD (Single Shot multi box Detector), etc. can be used in this step in principle. Since the YOLO and the SSD directly carry out regression prediction on the region where the target is located, and belong to one-state algorithms, the detection accuracy is usually sacrificed to obtain higher detection real-time performance. In the fingerprint positioning, the establishment of a fingerprint database has higher requirements on the position accuracy, and the link is allowed to be performed off-line, so that the dependence on the real-time performance of a target detection algorithm is reduced, and the adoption of the two-stage algorithm Faster RCNN with higher detection accuracy becomes a better choice. In addition, because the moving carrier has the problem of multi-scale target change caused by the wandering in the indoor environment, and currently, a plurality of improved methods for the problem exist, the invention recommends using fast RCNN with multi-scale target detection capability as the target detection model of the step.
And thirdly, analyzing the position information of the moving carrier.
The model trained in the second step can be used for carrying out target detection on the picture after the video is framed in the step to obtain the position of the trolley. However, the object detection model detects all pictures independently, and the large number of video frames will cause the detection process to have a relatively high time complexity. Considering the position dependence of the moving carrier in a short time, an object tracking algorithm may be introduced in this step: for example, the target is detected by fast RCNN at fixed time intervals, and the target is tracked by adopting a common Kernel Correlation filtering and tracking algorithm (KCF) in the time interval, so that the position Correlation of the target is fully utilized, the stability of the process is improved, and the calculation amount overhead is reduced.
Compared with the traditional target detection and tracking, the fingerprint indoor positioning scene mainly has two differences, and the invention provides targeted improvement based on the example:
(1) The indoor environment is complex and is easy to generate objects similar to the target object (moving carrier), which causes errors of the target object detected or tracked by the algorithm. The traditional scene tends to detect and track all target objects in the picture, and only a moving carrier carrying an antenna is a unique target in the fingerprint indoor positioning scene, so that in order to keep the tracking state of an original target when the fast RCNN detects the target position and prevent the tracking target from changing, the invention introduces a prior frame with the right for the fast RCNN algorithm: according to the method, after KCF tracking of the target is finished each time, the tracking result of the KCF is utilized to provide position prior for target detection, namely, pixel points near a last target area have higher prior weight, and the distribution form of the weight in space can be set by self (such as a two-dimensional Gaussian window), so that the attention to other peripheral objects can be effectively reduced, and non-target objects are well prevented from being detected or tracked.
(2) The construction of the fingerprint map library is carried out off-line, and the traditional target tracking algorithm is prone to real-time tracking. Future information can be fully utilized in the process of processing data offline. Specifically, the following steps are mainly performed for the target position with low reliability: in the process of target detection, the detection model outputs a position sequence and a confidence sequence of a target, an abnormal data point is detected by using an abnormal detection algorithm, historical information and future information can be simultaneously used when the current point is subjected to abnormal detection, and an abnormal result can be repaired by interpolation processing.
And fourthly, constructing an offline fingerprint map library.
First, CSI and location information are time synchronized. If the CSI acquiring device needs to perform resynchronization every other period of time and cannot output the CSI estimation value in the synchronization process, in this case, the time of each resynchronization needs to be recorded and output, and a resynchronization mark is used for achieving time synchronization of CSI information and location information in the data processing process.
Then, for the CSI smoothing processing, the CSI smoothing processing in the time domain and the frequency domain can effectively suppress noise. According to the invention, on the premise of small change of a CSI curve, a longer smooth window is selected on a time domain to realize noise suppression; the frequency domain then recommends setting a shorter smoothing window to avoid the CSI curve losing frequency domain details.
Then, the CSI sequence and the position sequence are sampled and matched. The sampling method with fixed time intervals has poor adaptability, because the CSI at certain positions changes faster or the speed of a moving carrier is higher at some moments, the CSI is sampled more densely in time, and no reference point exists around the point to be measured during online positioning. Also, sometimes CSI changes slowly over time, requiring a larger time sampling interval to avoid invalid and repeated CSI sampling, so the present invention samples based on whether the CSI curve similarity changes within a certain threshold range. Specifically, the invention performs time sampling based on CSI similarity, and the basic idea is: setting a CSI similarity threshold range (including an upper limit and a lower limit), assuming that the last sampling point is A, and then aiming to find the next sampling point: firstly, taking A as a current point, finding a next candidate sampling point B by the step size step by the A, if the CSI similarity between the A and the B is too high and is larger than the upper limit of a threshold value, discarding the candidate sampling point B, taking the B as a new current point, and increasing the step size from the B to continue trying; and if the similarity between the current point and the candidate sampling point is too low and is lower than the lower threshold, keeping the current point unchanged, and keeping step = step/2 to continuously search the next candidate sampling point. And continuously executing the operation until the similarity of the CSI is within the threshold range, wherein the sampling point meeting the condition is the next legal sampling point of the A.
And finally, extracting amplitude and phase information of the CSI from each CSI sampling point to construct a CSI fingerprint, and matching and storing the CSI fingerprint with the position coordinates at corresponding moments one by one to obtain a CSI fingerprint database.
And fifthly, positioning on line.
And (3) establishing a fingerprint map library according to the first step to the fourth step, and positioning the point to be detected on line: and searching out a reference point similar to the fingerprint of the point to be measured from the fingerprint library, and estimating the position of the point to be measured according to the position coordinates of the reference point.
The method provided by the invention has the advantages that the CSI of each position is collected by the moving carrier which walks in an indoor scene, the position of the moving carrier is identified through the trained target detection model, and then the CSI and the position of the moving carrier are matched to construct a fingerprint map library, so that manual measurement in the scene is not needed, a large number of personnel are avoided, and the method can quickly obtain a high-precision fingerprint library and realize the purpose of reducing the huge expense of fingerprint library construction.
Claims (4)
1. An indoor CSI fingerprint positioning method for fast off-line library establishment is characterized by comprising the following steps:
step 1: fixing a transmitter and a camera in a scene, enabling a moving carrier carrying a receiver to move in an indoor environment, acquiring CSI (channel state information) of each position, and acquiring a video image of the moving carrier by the camera; wherein, the CSI represents channel state information;
and 2, step: collecting and constructing a training sample, and training a target detection model off line;
the training sample is obtained by extracting frames of the video acquired according to the mode of the step 1 and calibrating the position of a motion carrier in the image;
the target detection model adopts a Faster RCNN model, the input of the model is a video image, and the model outputs the position of a moving carrier;
and 3, step 3: detecting a moving carrier in a video image by using a trained target detection model at fixed time intervals, and tracking the moving carrier in the time interval;
introducing a prior frame with a right to a target detection model, specifically: after target tracking is finished at each time interval, providing position prior for target detection by using a tracking result, wherein pixel points arranged near a target area at the last time have higher prior weight;
and 4, step 4: constructing an off-line fingerprint map library by the collected CSI of each position and the detected position of the motion carrier, and carrying out on-line positioning based on a fingerprint map route; wherein, the step of constructing the offline fingerprint map library comprises the following steps:
(4.1) carrying out time synchronization on the collected CSI and the position of the moving carrier;
(4.2) smoothing the CSI sequence;
(4.3) sampling the CSI sequence;
the time sampling is performed based on the CSI similarity, and specifically includes: setting a CSI similarity threshold range, setting the last sampling point as A, taking the A as a current point, finding the next candidate sampling point B through step length step by the A, and calculating the CSI similarity between the A and the B; if the similarity between the A and the B is higher than the upper limit of the threshold value, discarding the B, taking the B as a new current point, and continuously searching a next candidate sampling point by step length; if the similarity between A and B is lower than the lower threshold, setting step length to be halved, and searching the next candidate sampling point by taking A as the current point again; if the similarity of A and B is within the threshold range, taking B as the next legal sampling point of A;
and (4.4) extracting amplitude and phase information of the CSI from each CSI sampling point to construct a CSI fingerprint, and matching and storing the CSI fingerprint with the position coordinates at corresponding moments one by one to obtain an offline fingerprint map library.
2. The method according to claim 1, wherein in step 1, the camera takes a top view of the moving carrier; the moving carrier walks according to the preset regular movement track.
3. The method of claim 1, wherein in step 1, a metal baffle is disposed in the indoor scene to block the transmitting antenna, so as to create an NLOS environment.
4. The method according to claim 1, wherein in step 3, the abnormal data point detection is further performed on the position sequence and the confidence sequence of the motion carrier output by the target detection model, and when the current data point is detected to be abnormal, the historical information and the future information are utilized, and the abnormal data point is repaired by interpolation processing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111342171.9A CN114268918B (en) | 2021-11-12 | 2021-11-12 | Indoor CSI fingerprint positioning method for rapid off-line library building |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111342171.9A CN114268918B (en) | 2021-11-12 | 2021-11-12 | Indoor CSI fingerprint positioning method for rapid off-line library building |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114268918A CN114268918A (en) | 2022-04-01 |
| CN114268918B true CN114268918B (en) | 2022-10-18 |
Family
ID=80825182
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111342171.9A Active CN114268918B (en) | 2021-11-12 | 2021-11-12 | Indoor CSI fingerprint positioning method for rapid off-line library building |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114268918B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| USRE48004E1 (en) * | 2011-06-10 | 2020-05-19 | Lg Electronics Inc. | Method for transmitting control information and apparatus for same |
| CN112712557A (en) * | 2020-12-17 | 2021-04-27 | 上海交通大学 | Super-resolution CIR indoor fingerprint positioning method based on convolutional neural network |
| CN113343863A (en) * | 2021-06-11 | 2021-09-03 | 北京邮电大学 | Fusion characterization network model training method, fingerprint characterization method and equipment thereof |
| CN113518307A (en) * | 2021-04-22 | 2021-10-19 | 上海大学 | Position fingerprint positioning method based on CSI and crowdsourcing migration self-calibration updating |
-
2021
- 2021-11-12 CN CN202111342171.9A patent/CN114268918B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| USRE48004E1 (en) * | 2011-06-10 | 2020-05-19 | Lg Electronics Inc. | Method for transmitting control information and apparatus for same |
| CN112712557A (en) * | 2020-12-17 | 2021-04-27 | 上海交通大学 | Super-resolution CIR indoor fingerprint positioning method based on convolutional neural network |
| CN113518307A (en) * | 2021-04-22 | 2021-10-19 | 上海大学 | Position fingerprint positioning method based on CSI and crowdsourcing migration self-calibration updating |
| CN113343863A (en) * | 2021-06-11 | 2021-09-03 | 北京邮电大学 | Fusion characterization network model training method, fingerprint characterization method and equipment thereof |
Non-Patent Citations (2)
| Title |
|---|
| 《A Flexible 4G/5G Control Platform for Fingerprint-based Indoor Localization》;Michele Gucciardo;《IEEE INFOCOM 2019 - IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS)》;20190923;全文 * |
| 基于3DCNN的CSI-cluster室内指纹定位算法;李新春等;《重庆邮电大学学报(自然科学版)》;20200615(第03期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114268918A (en) | 2022-04-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108632761B (en) | Indoor positioning method based on particle filter algorithm | |
| CN110856112A (en) | Crowd-sourcing perception multi-source information fusion indoor positioning method and system | |
| CN101945284B (en) | Motion estimation device and method | |
| CN106952289B (en) | The WiFi object localization method analyzed in conjunction with deep video | |
| KR102110813B1 (en) | SLAM method and apparatus robust to wireless environment change | |
| CN104700408B (en) | A kind of indoor single goal localization method based on camera network | |
| CN107801203B (en) | Population density and mobility estimation method and system based on multi-data fusion | |
| CN107576325B (en) | A kind of indoor positioning terminal merging visual odometry and Magnetic Sensor | |
| CN111277946A (en) | Fingerprint database self-adaptive updating method in Bluetooth indoor positioning system | |
| Wang et al. | RFID & vision based indoor positioning and identification system | |
| CN112529962A (en) | Indoor space key positioning technical method based on visual algorithm | |
| Nguyen et al. | Low speed vehicle localization using wifi fingerprinting | |
| Feng et al. | Visual map construction using RGB-D sensors for image-based localization in indoor environments | |
| He et al. | WiFi iLocate: WiFi based indoor localization for smartphone | |
| CN112164093A (en) | Automatic person tracking method based on edge features and related filtering | |
| Quezada-Gaibor et al. | Data cleansing for indoor positioning Wi-Fi fingerprinting datasets | |
| CN111308420A (en) | Indoor non-line-of-sight positioning method based on acoustic signal time delay estimation and arrival frequency | |
| Wei et al. | MM-Loc: Cross-sensor indoor smartphone location tracking using multimodal deep neural networks | |
| CN114268918B (en) | Indoor CSI fingerprint positioning method for rapid off-line library building | |
| KR100994722B1 (en) | Method for tracking moving object on multiple cameras using probabilistic camera hand-off | |
| CN108932478A (en) | Object positioning method, device and shopping cart based on image | |
| Sun et al. | Device-free human localization using panoramic camera and indoor map | |
| CN115808170B (en) | Indoor real-time positioning method integrating Bluetooth and video analysis | |
| CN109001674B (en) | WiFi fingerprint information rapid acquisition and positioning method based on continuous video sequence | |
| TWI425241B (en) | Combining the signal intensity characteristic comparison and position prediction analysis of hybrid indoor positioning method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |