WO2021093872A1 - 一种群智感知的多源信息融合室内定位方法及系统 - Google Patents

一种群智感知的多源信息融合室内定位方法及系统 Download PDF

Info

Publication number
WO2021093872A1
WO2021093872A1 PCT/CN2020/128802 CN2020128802W WO2021093872A1 WO 2021093872 A1 WO2021093872 A1 WO 2021093872A1 CN 2020128802 W CN2020128802 W CN 2020128802W WO 2021093872 A1 WO2021093872 A1 WO 2021093872A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
wifi
target
fingerprint
algorithm
Prior art date
Application number
PCT/CN2020/128802
Other languages
English (en)
French (fr)
Inventor
赵毓斌
李芳敏
须成忠
Original Assignee
深圳先进技术研究院
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 深圳先进技术研究院 filed Critical 深圳先进技术研究院
Publication of WO2021093872A1 publication Critical patent/WO2021093872A1/zh

Links

Images

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04WWIRELESS COMMUNICATION NETWORKS
    • H04W4/00Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/02Services making use of location information
    • H04W4/021Services related to particular areas, e.g. point of interest [POI] services, venue services or geofences
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01CMEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/005Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 with correlation of navigation data from several sources, e.g. map or contour matching
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01CMEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/20Instruments for performing navigational calculations
    • G01C21/206Instruments for performing navigational calculations specially adapted for indoor navigation
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01SRADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S5/00Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations
    • G01S5/02Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations using radio waves
    • G01S5/0257Hybrid positioning
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04WWIRELESS COMMUNICATION NETWORKS
    • H04W4/00Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/30Services specially adapted for particular environments, situations or purposes
    • H04W4/33Services specially adapted for particular environments, situations or purposes for indoor environments, e.g. buildings

Definitions

  • the present invention relates to the field of positioning, and in particular to a multi-source information fusion indoor positioning method and system for group intelligence perception.
  • GPS global positioning system
  • the current indoor positioning solutions can be divided into two categories: one is the indoor positioning algorithm based on the model, and the other is the positioning algorithm based on the fingerprint database.
  • Model-based indoor positioning algorithms are limited by the quality of the model and the errors caused by signal measurement, while indoor positioning algorithms based on fingerprint libraries can avoid this problem.
  • the positioning algorithm based on the fingerprint database is mainly divided into two parts: one is the offline sampling phase, and the other is the online positioning part.
  • the offline sampling stage mainly collects the information of indoor sensors used for positioning, such as geomagnetic intensity information, Bluetooth, WiFi signal strength values, and images.
  • the online stage mainly collects the geomagnetism, Bluetooth, WiFi, and pictures of the target location. The information is matched with the information in the fingerprint library to determine the location of the target.
  • fusion of heterogeneous information to enhance position estimation is a main effective method, that is, to establish a fingerprint database of indoor fusion information for positioning.
  • fingerprint information fusion the main problem based on fingerprint information fusion is the need to collect a large amount of data to build a reliable database, and the establishment of such a database consumes a lot of manpower and time.
  • crowd perception is an effective solution for database construction.
  • mobile users can share different location-related information from their smartphones, such as WiFi, geomagnetism, or visual images, so that the positioning system can derive accurate location estimates based on this information.
  • Some visual images can directly lead to precise locations, and the geomagnetic data is quite stable.
  • WiFi data can also be obtained for free whether you are walking, running or standing still.
  • Existing indoor positioning algorithms include smart phone indoor navigation methods based on hybrid WiFi, geomagnetism and dead reckoning (PDR), AP-weighted multiple matching nearest neighbor method for fingerprint-based indoor positioning, and use of ubiquitous geomagnetism and WiFi anomalies
  • PDR geomagnetism and dead reckoning
  • the Magicol positioning method system for point positioning there are positioning methods that use deep learning to identify indoor positioning fingerprint libraries
  • VMag systems that use geomagnetism and image sensors to locate, and perform complex neural network training on image information, and only use images
  • An algorithm for visual positioning This method requires the establishment of a large amount of image fingerprint library information and matches with the characteristics of each frame of the image, so it takes more time.
  • the embodiment of the present invention provides a multi-source information fusion indoor positioning method and system with intelligent perception of the population, so as to at least solve the technical problem of low accuracy of the existing indoor positioning method.
  • a multi-source information fusion indoor positioning method with intelligent perception of the population which includes the following steps:
  • the image-based small area determination algorithm IBSM and the segmented structure adaptive algorithm are used in locating the target location.
  • SSAC to estimate and adaptively calibrate the target position.
  • constructing a fingerprint recognition database based on group perception includes: constructing image fingerprint recognition and using a dynamic trajectory collection method to construct a fingerprint recognition database.
  • the fingerprint recognition database includes the geomagnetic intensity and the RSS value of WiFi.
  • constructing a fingerprint recognition database based on group perception includes:
  • Establishment of the visual image of the characteristic area build an image database and link each image with the relevant location; when the target tries to self-calibrate, search for visual information and match it with the image database;
  • WiFi and geomagnetic dynamic fingerprint database use dynamic trajectory collection to build WiFi RSS and geomagnetic intensity database.
  • the user walks the smart phone indoors along a predetermined barrier-free path and uploads the data.
  • the RSS data of each location and WiFi The geomagnetic intensity is acquired synchronously; after constructing the entire indoor data collection, an indoor dynamic path diagram is formed.
  • fusing the collected geomagnetic, WiFi and visual image information and locating the target location based on the constructed fingerprint recognition database includes:
  • the image-based sub-region matching method IBSM maps each sampled image to the designated region determined when the image fingerprint is created;
  • the segmented structure adaptive calibration is performed on the WiFi and geomagnetic fingerprint data acquired simultaneously, and after evaluating the similarity between the geomagnetic sequence obtained in the online phase and the geomagnetic sequence in fingerprint identification, the matching of the two sequences is determined.
  • SIFT feature descriptor includes:
  • the part around the key point is divided into multiple concentric circles.
  • the radii are one pixel, two pixels, and even multiple pixels.
  • using a circle to construct the SIFT feature descriptor also includes: using a random sample consensus method RANSAC to eliminate mismatch points.
  • the image-based sub-region matching method IBSM corresponds to each sampled image to the specified region determined when the image fingerprint is created, including: when the target is close to a certain sampling point, the sub-region matching method IBSM will directly give a special The location coordinates or sub-area range of the location, and then use the WKNN algorithm to measure the precise initial location.
  • segmented structure adaptive calibration of the WiFi and geomagnetic fingerprint data acquired simultaneously includes: using the dynamic time warping algorithm DTW to treat the geomagnetic intensity sequence as a continuous waveform, and find a matching point in the database.
  • performing segment structure adaptive calibration on the fingerprint data of WiFi and geomagnetism acquired simultaneously also includes: the geomagnetism sequence is segmented, the position will be updated after calibration, and then a new matching calibration will be started.
  • a multi-source information fusion indoor positioning system with intelligent perception of the population including:
  • Fingerprint identification database construction unit used to build fingerprint identification database based on group perception
  • the positioning unit is used to fuse the collected geomagnetic, WiFi and visual image information and locate the target location based on the constructed fingerprint recognition database.
  • the image-based small area determination algorithm IBSM and analysis are used to locate the target location.
  • Segment structure adaptive algorithm SSAC to estimate and adaptively calibrate the target position.
  • the multi-source information fusion indoor positioning method and system for group intelligence perception in the embodiments of the present invention adopts a fingerprint recognition technology based on group perception, which has significant effects, especially for complex infrastructures.
  • the present invention collects three free indoor resource information, namely geomagnetism, WiFi, and visual images. Since the signal strength of a single information source is unstable, the present invention combines these three information sources. Perform fusion to estimate the position of the target.
  • the present invention also proposes two new algorithms: image-based small area determination algorithm (IBSM) and segmented structure adaptive algorithm (SSAC), which can help the target quickly determine the indoor small area range or determine The special location of the target and help the target determine the accuracy of positioning or the range of the calibration target.
  • IBSM image-based small area determination algorithm
  • SCA segmented structure adaptive algorithm
  • Fig. 1 is a flowchart of a multi-source information fusion indoor positioning method based on group intelligence perception of the present invention
  • FIG. 2 is a frame diagram of the Wimage system in the present invention.
  • Figure 3 is a distribution diagram of indoor image sampling points in the present invention.
  • Figure 4 is a vector diagram of the descriptor in the present invention.
  • Fig. 5 is a graph of geomagnetic data of different walking speeds on the same path in the present invention.
  • Figure 6 is a graph of geomagnetic data with different mobile phone postures on the same path in the present invention.
  • Figure 7 is a performance comparison diagram of IBSM and SSAC in the Wimage system of the present invention.
  • FIG. 8 is a diagram of WiFi signal strength in the present invention.
  • Fig. 9 is a performance comparison diagram of the geomagnetic correction of the present invention in SSAC.
  • Figure 10 is an effect diagram of the indoor tracking image of the Wimage system in the present invention.
  • Fig. 11 is a block diagram of a multi-source information fusion indoor positioning system for group intelligence perception according to the present invention.
  • the present invention constructs an effective framework Wimage, which combines visual images, WiFi and geomagnetism as a comprehensive fingerprint database, and integrates these heterogeneous data according to their characteristics to derive the target position. And image vision, geomagnetism, and WiFi information are completely free indoors, and users can obtain this information for free, and no additional equipment is required.
  • Wimage visual image data are reference pictures that indicate some key features related to certain specific locations, such as room numbers, kitchens, and fire hydrants. These collected images can use the matching algorithm Scale Invariant Feature Transformation (SIFT) to directly indicate a special location or positioning range, which helps the system reduce coarse-grained positioning errors.
  • SIFT Scale Invariant Feature Transformation
  • WiFi and geomagnetic data through crowd perception, and locates the target location through positioning methods. These methods include weighted k-nearest neighbor (WKNN) and dynamic time warping (DTW). Specifically, it uses WiFi data based on the WKNN algorithm to estimate the target position, and the geomagnetic data of the DTW algorithm is used for calibration and correction through segmented geomagnetic sequence matching. In this case, the target position can be corrected automatically.
  • WKNN weighted k-nearest neighbor
  • DTW dynamic time warping
  • the Wimage system proposed by the present invention uses a crowd perception method to construct a positioning fingerprint library and integrates three types of information: image vision, geomagnetism and WiFi to design a positioning algorithm.
  • Fingerprint-based indoor positioning algorithm In order to ensure the accuracy of positioning when establishing the fingerprint database, a large number of relevant data needs to be collected at multiple indoor sampling points, where the sampling points are based on the overall indoor environment and conditions to grid the indoor area, etc. To determine, the sparseness or denseness of the grid points greatly affects the accuracy of indoor positioning. At the same time, it is necessary to use interpolation algorithms to determine fingerprint information for unsampled points, which will not only bring information inaccuracy but also cost a lot of The cost of manpower, material resources and time is required for the staff to collect relevant information one by one in the well-divided sampling points in the room when establishing the fingerprint database. This requires a lot of work in the process of establishing the fingerprint database in the early stage.
  • the present invention adopts a crowd-sensing method in the Wimage system to encourage different mobile users to share their perception data and upload it to the server.
  • This method can use different users and different devices to detect when the fingerprint database is established.
  • An indoor fingerprint library is established based on the information of the system.
  • the present invention adopts a method different from the traditional discrete point library building, but adopts a route trajectory collection method.
  • the staff can hold the smart phone along the indoor feasibility
  • the route synchronously collects geomagnetism and WiFi information, which greatly reduces the establishment time of the fingerprint database, because the staff only need to walk along the trajectory of the indoor route to collect the indoor geomagnetism and WiFi information.
  • the fingerprint database established by this method is also There is no need to resort to other interpolation algorithms to calculate the fingerprint library information of the unsampled points.
  • the Wimage system of the present invention uses indoor WiFi, geomagnetism, and images for positioning.
  • the use of these three indoor resources for positioning not only does not require additional deployment indoors, but is also completely feasible.
  • image resources are available everywhere, and different pictures have different characteristics, and different picture characteristics can be directed to different indoor locations, which provides the possibility to use pictures for positioning.
  • the earth itself is a huge magnetic field, so geomagnetic information also exists everywhere in the room, and the geomagnetic field information is different in different locations, so different geomagnetic information can be used to locate.
  • WiFi has become more and more popular indoors.
  • WiFi In office buildings, large hospitals, and large conference venues, WiFi is basically available, and the signal strength of WiFi increases with the distance from the transmitting port. The value is different, and a certain location will receive different WiFi signal strength (RSS) values of different WiFi ports, so the WiFi information received at each location in the room is different, based on this feature WiFi RSS can be used To perform indoor positioning.
  • RSS WiFi signal strength
  • the fingerprint-based indoor positioning algorithm needs to rely on a certain positioning technology such as Bluetooth, WiFi and infrared, but a single signal source will have signal fluctuations and unstable due to noise or obstacles, so a single signal is used It is difficult to accurately estimate the location of the target when the source is located.
  • the invention adopts three information sources of image, WiFi and geomagnetism for positioning, and proposes a reliable positioning algorithm, which can not only quickly locate and track the initial position of the target, but also correct the target position in time and accurately.
  • the specific method is: use indoor special pictures to specifically determine the location or range of the target.
  • the present invention uses an improved SIFT algorithm to match pictures, and proposes a small picture-based design for the entire process.
  • Area determination algorithm (IBSM) and then use the weighted K nearest neighbor algorithm to estimate the initial position of the target according to the WiFi signal strength RSSI, while tracking the target, use the trajectory change characteristics of the geomagnetic information to make phase corrections to the target location
  • the method used here is the dynamic time warping algorithm (DTW).
  • phase correction here is because the geomagnetic changes in a section of the path are regular, and the process of establishing the geomagnetic fingerprint database is also phased according to different paths. of. Aiming at the correction process of the target position, the present invention designs and proposes a segmented structure adaptive calibration (SSAC) algorithm, and experiments show that it can effectively improve the positioning accuracy.
  • SAC segmented structure adaptive calibration
  • the present invention proposes a Wimage positioning system.
  • the Wimage positioning system of the present invention adopts a fingerprint recognition technology based on group perception, which has significant effects, especially for complex infrastructure.
  • the present invention collects three free indoor resource information, namely geomagnetism, WiFi, and visual images. Since the signal strength of a single information source is unstable, the present invention combines these three information sources. Perform fusion to estimate the position of the target.
  • the present invention also proposes two new algorithms: image-based small area determination algorithm (IBSM) and segmented structure adaptive algorithm (SSAC), which can help the target quickly determine the indoor small area range or Determine the special location of the target and help the target determine the accuracy of positioning or calibrate the range of the target.
  • IBSM image-based small area determination algorithm
  • SCA segmented structure adaptive algorithm
  • the present invention provides a multi-source information fusion indoor positioning method and system based on the group intelligence perception of a positioning framework and algorithm using free indoor resources for position estimation.
  • the method and the system propose a method of using crowd perception to establish a fingerprint database.
  • the IBSM and SSAC algorithms are proposed in the paper, which can effectively reduce the positioning cost and the use of human and material resources, and can effectively improve the indoor positioning accuracy, and has a strong anti-noise and anti-interference ability.
  • a multi-source information fusion indoor positioning method with intelligent perception of the population is provided.
  • the method includes the following steps:
  • S102 Combine the collected geomagnetic, WiFi, and visual image information and locate the target location based on the constructed fingerprint recognition database, where the image-based small area determination algorithm IBSM and the segmented structure self are used to locate the target location.
  • the adaptive algorithm SSAC estimates and adaptively calibrate the target position.
  • the multi-source information fusion indoor positioning method for group intelligence perception in the embodiment of the present invention adopts a fingerprint recognition technology based on group perception, which has significant effects, especially for complex infrastructure.
  • the present invention collects three free indoor resource information, namely geomagnetism, WiFi, and visual images. Since the signal strength of a single information source is unstable, the present invention combines these three information sources. Perform fusion to estimate the position of the target.
  • the present invention also proposes two new algorithms: image-based small area determination algorithm (IBSM) and segmented structure adaptive algorithm (SSAC), which can help the target quickly determine the indoor small area range or determine The special location of the target and help the target determine the accuracy of positioning or the range of the calibration target.
  • IBSM image-based small area determination algorithm
  • SCA segmented structure adaptive algorithm
  • the present invention proposes a complete set of indoor positioning system Wimage, which includes two stages, namely the crowd perception stage and the target tracking stage.
  • the present invention constructs image fingerprint recognition and uses a dynamic trajectory collection method to construct a fingerprint recognition database including the geomagnetic intensity and the RSS value of WiFi.
  • the target tracking stage the target performs multi-data fusion positioning based on the collected images, geomagnetic measurements and WiFi signals.
  • a variety of methods are applied, including improved SIFT algorithm, WKNN and DTW algorithm for determining sub-regions through image matching to estimate and adaptively calibrate the target position. Specifically:
  • the images taken by the user are highly related to certain specific locations or sub-regions.
  • the present invention can use these items as the semantic information of the characteristic area.
  • the target tries to self-calibrate, it can use the camera to search for visual information and match it with an image database. Therefore, the present invention constructs an image database and links each image with a related location.
  • the present invention provides a collection point and location of indoor pictures. These pictures are fire hydrant, rest room, house number, etc.
  • the Wimage system uses dynamic trajectory collection to build WiFi RSS and geomagnetic intensity databases. Users only need to walk their smartphones indoors along a predetermined barrier-free path and upload their data, which makes it more convenient for users.
  • the RSS data and geomagnetic intensity of each location WiFi are obtained synchronously, so their information is also related to each other. After constructing the collected data of the entire indoor, the indoor dynamic path diagram is formed.
  • the image-based SIFT matching algorithm is an algorithm for extracting local features, searching for extreme points in the scale space, spatial scale and rotation invariants.
  • the SIFT algorithm is suitable for fast and accurate matching in massive databases.
  • Image matching is based on the extracted SIFT features, and the SIFT algorithm uses key points to describe image features.
  • the key point is the vector of floating numbers.
  • the length of the vector is determined by the descriptor.
  • the descriptor contains the direction and size of the direction histogram array.
  • a vector of descriptors is composed of a 4 ⁇ 4 array and 8 directions, as shown in Figure 4.
  • the left picture is a typical SIFT algorithm
  • the right picture is an improved SIFT algorithm.
  • R i (r i1, r i2, ... r i128) is an image matching SIFT descriptor vector.
  • S j (s j1 , s j2 ,...s j128 ) is the SIFT descriptor vector of the matched image that extracts N key points, and j ⁇ N.
  • Image matching is actually based on the measurement of Euclidean distance.
  • min ⁇ Dis ⁇ obtained from R i and S m in the present invention is the nearest Euclidean distance
  • SecMin ⁇ Dis ⁇ obtained from R i and S t is the second nearest neighbor Euclidean distance. They are respectively:
  • the improved SIFT algorithm is mainly to reduce the calculation cost and improve the matching speed.
  • the dimension of the descriptor is higher.
  • the image matching task in the Wimage system it will perform better in the low-dimensional descriptor.
  • the images in fingerprint recognition are usually very different. Therefore, in the improved SIFT, the present invention considers reducing the vector dimension of the descriptor to improve the matching speed, and uses a circle to construct the SIFT feature descriptor. Taking the key point as the center, the part around the key point is divided into 6 concentric circles. There are three concentric circles with different radii from the inside to the outside. The radii are 1 pixel, two pixels and three pixels. It is represented by different symbol lines in Figure 8.
  • R i (r i1, r i2, ... r i48)
  • S j (s j1, s j2, ... s j48).
  • the present invention uses Random Sample Consensus (RANSAC) to eliminate mismatch points to ensure matching accuracy.
  • R i (r i1, r i2, ... r i48)
  • S j (s j1, s j2, ... s j48).
  • each sampled image corresponds to a designated area determined when the image fingerprint is created.
  • Each sampled image P will correspond to the coordinate range [X, Y], where X represents the range of [x, x'], and Y represents the range of [y, y'].
  • Figure 3 shows the relationship between the picture and the coordinates of the sub-area.
  • Image matching is based on the improved SIFT feature. If two similar images P 0 and P 1 meet the threshold range of formula (3), then the sub-region of the target will be quickly locked in the range corresponding to the successfully matched candidate image P 1 [ X 1 ,Y 1 ].
  • the present invention defines this algorithm as an image-based subregion matching (IBSM) method.
  • IBSM image-based subregion matching
  • the image in fingerprint recognition contains detailed semantic information related to location.
  • the present invention uses an improved SIFT algorithm, which is feasible.
  • the target can use image matching to determine the location sub-region. When the target is close to a certain sampling point, it can directly give a special position coordinate or sub-area range, and then measure the precise initial position through the WKNN algorithm.
  • the matching algorithm of the WiFi fingerprint database adopts the weighted K-nearest neighbor (WKNN) method, which evaluates the relationship between the RSS value of each WiFi wireless access point (AP) collected by the mobile terminal and the RSS value of each AP corresponding to the sampling point in the fingerprint. Similarity. (The measurement of similarity generally adopts Euclidean distance) Determine the positions of k sampling points with high similarity, and use the weighted average to derive the user's position:
  • the composition of each track fingerprint database is: Where t is the length of the h-th sampled track.
  • the AP signal strength collected at sampling point j in the h-th sampled trajectory is among them Is the RSS value of the nth AP collected at the sampling point j at the hth sampling trajectory.
  • the variance reflects the dispersion of the sample data distribution. The greater the variance, the greater the fluctuation of the rss value, so the weight expression is:
  • the present invention adopts geomagnetic calibration.
  • the fingerprint data of WiFi and geomagnetism are obtained synchronously, which provides the feasibility of geomagnetism calibration.
  • the invention uses dynamic track collection to construct a geomagnetic fingerprint library. Therefore, geomagnetic data can be considered as multiple sequence segments.
  • the present invention knows that different users have different walking speeds, and the shapes of geomagnetic waves collected on the same path are similar, as shown in FIG. 5 and FIG. 6.
  • the dynamic time warping algorithm (DTW) is a method to measure the similarity of two time series of different lengths. It can dynamically match the points of two similar waveforms.
  • DTW treats the geomagnetic intensity sequence as a continuous waveform and finds a matching point in the database.
  • Z and X need to be extended or shortened to align in shape.
  • the planning path W represents this mapping relationship:
  • the least cost path can also be expressed as:
  • Geomagnetic calibration is of great significance to the accuracy of positioning. It can check the result of WiFi positioning, confirm the accuracy of the positioning result or correct the wrong positioning. Considering this function, the present invention matches geomagnetic data in subdivisions. In addition, although the use of a single geomagnetic data for positioning will be affected by noise, it is more reliable for sequence segment matching results. This method is defined as an algorithm called segmented structure adaptive calibration (SSAC). In addition, the geomagnetic sequence is segmented, the position will be updated after calibration, and then a new matching calibration will start, which can also avoid accumulated errors.
  • SAC segmented structure adaptive calibration
  • a multi-source information fusion indoor positioning system with intelligent perception of the population is provided, referring to Fig. 11, including:
  • the fingerprint recognition database construction unit 100 is used to construct a fingerprint recognition database based on group perception;
  • the positioning unit 200 is used to fuse the collected geomagnetic, WiFi, and visual image information and locate the target location based on the constructed fingerprint recognition database.
  • the image-based small area determination algorithm IBSM and IBSM are used to locate the target location.
  • the segmented structure adaptive algorithm SSAC is used to estimate and adaptively calibrate the target position.
  • the multi-source information fusion indoor positioning system for group intelligence perception in the embodiment of the present invention adopts a fingerprint recognition technology based on group perception, which has significant effects, especially for complex infrastructure.
  • the present invention collects three free indoor resource information, namely geomagnetism, WiFi, and visual images. Since the signal strength of a single information source is unstable, the present invention combines these three information sources. Perform fusion to estimate the position of the target.
  • the present invention also proposes two new algorithms: image-based small area determination algorithm (IBSM) and segmented structure adaptive algorithm (SSAC), which can help the target quickly determine the indoor small area range or determine The special location of the target and help the target determine the accuracy of positioning or the range of the calibration target.
  • IBSM image-based small area determination algorithm
  • SCA segmented structure adaptive algorithm
  • the present invention proposes a complete set of indoor positioning system Wimage, which includes two stages, namely the crowd perception stage and the target tracking stage.
  • the present invention constructs image fingerprint recognition and uses a dynamic trajectory collection method to construct a fingerprint recognition database including the geomagnetic intensity and the RSS value of WiFi.
  • the target tracking stage the target performs multi-data fusion positioning based on the collected images, geomagnetic measurements and WiFi signals.
  • a variety of methods are applied, including improved SIFT algorithm, WKNN and DTW algorithm for determining sub-regions through image matching to estimate and adaptively calibrate the target position. Specifically:
  • Fingerprint recognition database construction unit 100 database establishment based on crowd perception
  • the images taken by the user are highly related to certain specific locations or sub-regions.
  • the present invention can use these items as the semantic information of the characteristic area.
  • the target tries to self-calibrate, it can use the camera to search for visual information and match it with the image database. Therefore, the present invention constructs an image database and links each image with a related location.
  • the present invention provides a collection point and location of indoor pictures. These pictures are fire hydrant, rest room, house number, etc.
  • the Wimage system uses dynamic trajectory collection to build WiFi RSS and geomagnetic intensity databases. Users only need to walk their smartphones indoors along a predetermined barrier-free path and upload their data, which makes it more convenient for users.
  • the RSS data and geomagnetic intensity of each location WiFi are obtained synchronously, so their information is also related to each other. After constructing the collected data of the entire indoor, the indoor dynamic path diagram is formed.
  • Positioning unit 200 target positioning and tracking algorithm
  • the image-based SIFT matching algorithm is an algorithm for extracting local features, searching for extreme points in the scale space, spatial scale and rotation invariants.
  • the SIFT algorithm is suitable for fast and accurate matching in massive databases.
  • Image matching is based on the extracted SIFT features, and the SIFT algorithm uses key points to describe image features.
  • the key point is the vector of floating numbers.
  • the length of the vector is determined by the descriptor.
  • the descriptor contains the direction and size of the direction histogram array.
  • a vector of descriptors is composed of a 4 ⁇ 4 array and 8 directions, as shown in Figure 4.
  • the left picture is a typical SIFT algorithm
  • the right picture is an improved SIFT algorithm.
  • R i (r i1, r i2, ... r i128) is an image matching SIFT descriptor vector.
  • S j (s j1 , s j2 ,...s j128 ) is the SIFT descriptor vector of the matched image that extracts N key points, and j ⁇ N.
  • Image matching is actually based on the measurement of Euclidean distance.
  • min ⁇ Dis ⁇ obtained from R i and S m in the present invention is the nearest Euclidean distance
  • SecMin ⁇ Dis ⁇ obtained from R i and S t is the second nearest neighbor Euclidean distance. They are respectively:
  • the improved SIFT algorithm is mainly to reduce the calculation cost and improve the matching speed.
  • the dimension of the descriptor is higher.
  • the image matching task in the Wimage system it will perform better in the low-dimensional descriptor.
  • the images in fingerprint recognition are usually very different. Therefore, in the improved SIFT, the present invention considers reducing the vector dimension of the descriptor to improve the matching speed, and uses a circle to construct the SIFT feature descriptor. Taking the key point as the center, the part around the key point is divided into 6 concentric circles. There are three concentric circles with different radii from the inside to the outside. The radii are 1 pixel, two pixels and three pixels. It is represented by different symbol lines in Figure 8.
  • R i (r i1, r i2, ... r i48)
  • S j (s j1, s j2, ... s j48).
  • the present invention uses Random Sample Consensus (RANSAC) to eliminate mismatch points to ensure matching accuracy.
  • R i (r i1, r i2, ... r i48)
  • S j (s j1, s j2, ... s j48).
  • each sampled image corresponds to a designated area determined when the image fingerprint is created.
  • Each sampled image P will correspond to the coordinate range [X, Y], where X represents the range of [x, x'], and Y represents the range of [y, y'].
  • Figure 3 shows the relationship between the picture and the coordinates of the sub-area.
  • Image matching is based on the improved SIFT feature. If two similar images P 0 and P 1 meet the threshold range of formula (3), then the sub-region of the target will be quickly locked in the range corresponding to the successfully matched candidate image P 1 [ X 1 ,Y 1 ].
  • the present invention defines this algorithm as an image-based subregion matching (IBSM) method.
  • IBSM image-based subregion matching
  • the image in fingerprint recognition contains detailed semantic information related to location.
  • the present invention uses an improved SIFT algorithm, which is feasible.
  • the target can use image matching to determine the location sub-region. When the target is close to a certain sampling point, it can directly give a special position coordinate or sub-area range, and then measure the precise initial position through the WKNN algorithm.
  • the matching algorithm of the WiFi fingerprint database adopts the weighted K-nearest neighbor (WKNN) method, which evaluates the relationship between the RSS value of each WiFi wireless access point (AP) collected by the mobile terminal and the RSS value of each AP corresponding to the sampling point in the fingerprint. Similarity. (The measurement of similarity generally adopts Euclidean distance) Determine the positions of k sampling points with high similarity, and use the weighted average to derive the user's position:
  • the composition of each track fingerprint database is: Where t is the length of the h-th sampled track.
  • the AP signal strength collected at sampling point j in the h-th sampled trajectory is among them Is the RSS value of the nth AP collected at the sampling point j at the hth sampling trajectory.
  • the variance reflects the dispersion of the sample data distribution. The greater the variance, the greater the fluctuation of the rss value, so the weight expression is:
  • the present invention adopts geomagnetic calibration.
  • the fingerprint data of WiFi and geomagnetism are obtained synchronously, which provides the feasibility of geomagnetism calibration.
  • the invention uses dynamic track collection to construct a geomagnetic fingerprint library. Therefore, geomagnetic data can be considered as multiple sequence segments.
  • the present invention knows that different users have different walking speeds, and the shapes of geomagnetic waves collected on the same path are similar, as shown in FIG. 5 and FIG. 6.
  • the dynamic time warping algorithm (DTW) is a method to measure the similarity of two time series of different lengths. It can dynamically match the points of two similar waveforms.
  • DTW treats the geomagnetic intensity sequence as a continuous waveform and finds a matching point in the database.
  • Z and X need to be extended or shortened to align in shape.
  • the planning path W represents this mapping relationship:
  • the least cost path can also be expressed as:
  • Geomagnetic calibration is of great significance to the accuracy of positioning. It can check the result of WiFi positioning, confirm the accuracy of the positioning result or correct the wrong positioning. Considering this function, the present invention matches geomagnetic data in subdivisions. In addition, although the use of a single geomagnetic data for positioning will be affected by noise, it is more reliable for sequence segment matching results. This method is defined as an algorithm called segmented structure adaptive calibration (SSAC). In addition, the geomagnetic sequence is segmented, the position will be updated after calibration, and then a new matching calibration will start, which can also avoid accumulated errors.
  • SAC segmented structure adaptive calibration
  • the innovation of the present invention is at least:
  • the present invention proposes an improved SIFT algorithm.
  • the improved SIFT algorithm simplifies the feature descriptor of the image, and at the same time ensures the accuracy of image matching in the Wimage system, and the simplified feature descriptor is extremely large This reduces computing resources and costs, and at the same time improves the efficiency of image matching;
  • the multi-source information selected by the present invention during positioning is WiFi, geomagnetism, and images, and these three types of information are freely available indoors;
  • the present invention uses a crowd-sensing method to save staff time and material resources for establishing a fingerprint database, and when collecting the fingerprint database, the present invention proposes to use trajectory planning to synchronize Collect WiFi and geomagnetic information. This method only needs to walk along the indoor path to obtain the indoor WiFi signal strength and geomagnetic strength value, which simplifies the construction of the fingerprint library and improves the accuracy of the fingerprint library;
  • the present invention proposes an adaptive calibration scheme using geomagnetism for segmented structure, which can correct the position of the target in time and greatly improve the positioning accuracy.
  • the Wimage proposed by the present invention is evaluated in a comprehensive experiment. Constructed including indoor landmarks, RSS database, geomagnetic intensity database and visual image database. In the crowd perception stage, all RSS and geomagnetic intensity data are collected through dynamic trajectory collection. The total number of APs in the experiment of the present invention is 26, and the scan period of WiFi is set to 0.02s.
  • the role of IBSM in Wimage is mainly to reduce the positioning range in the initial positioning and accurately divide the target into small areas. Determining the location or sub-region of the sampling point mainly uses image matching.
  • the invention adopts the improved SIFT, which greatly reduces the calculation cost when the matching accuracy is satisfied. As shown in Figure 7, the present invention can see from the accumulated error that the IBSM can help the target improve the positioning accuracy.
  • the WKNN algorithm uses WiFi data to locate the target, but due to environmental influences, the WiFi data is very unstable.
  • the present invention randomly tracks the signal strength of three WiFi ports on a path, and Figure 8 reflects the fluctuation of WiFi strength. Therefore, the accuracy of positioning will be greatly affected.
  • the present invention evaluates the performance of the geomagnetic calibration. As shown in Fig. 9, in the SSAC algorithm, the geomagnetic calibration improves the positioning accuracy.
  • the present invention uses the Wimage system to track a route in an office building. As shown in Fig. 10, the solid line represents the actual route, and the dotted and dashed line represents the estimated route. It can be clearly seen that the estimated route is close to the actual route.
  • the present invention measures IBSM and SSAC algorithms in Wimage. As can be seen in Fig. 7, the present invention has a great influence on the accuracy of the target position. In addition, it is very convenient to correct geomagnetism data, because geomagnetism is everywhere.
  • the improved SIFT algorithm in the present invention can also directly use the SIFT algorithm to match pictures, but it will increase computing resources and time; the solution designed by the present invention can be applied not only in office buildings but also in hospitals, large conference venues, shopping malls, etc. Application in the scene.
  • the disclosed technical content can be implemented in other ways.
  • the system embodiment described above is only illustrative.
  • the division of units may be a logical function division, and there may be other divisions in actual implementation.
  • multiple units or components may be combined or integrated into Another system, or some features can be ignored, or not implemented.
  • the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, units or modules, and may be in electrical or other forms.
  • the units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the objectives of the solutions of the embodiments.
  • the functional units in the various embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit.
  • the above-mentioned integrated unit can be implemented in the form of hardware or software functional unit.
  • the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium.
  • the technical solution of the present invention essentially or the part that contributes to the existing technology or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium.
  • a computer device which may be a personal computer, a server, or a network device, etc.
  • the aforementioned storage media include: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program code .

Landscapes

  • Engineering & Computer Science (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Image Analysis (AREA)

Abstract

本发明涉及一种群智感知的多源信息融合室内定位方法及系统。首先基于群体感知构建指纹识别数据库,再将采集到的地磁、WiFi和视觉图像信息进行融合并基于构建的指纹识别数据库对目标位置进行定位,其中在对目标位置进行定位中使用基于图像的小区域确定算法IBSM和分段结构自适应算法SSAC来估计和自适应地校准目标位置。该方法及系统采用了一种基于群体感知的指纹识别技术,该技术效果显著。由于单个信息源的信号强度是不稳定的,因此本发明将这三种信息源进行融合来对目标的位置进行估计。本发明中的两个算法能够帮助目标快速确定所在室内的小区域范围或者确定目标所在的特殊位置以及帮助目标确定定位的准确性或者校准目标所在范围。

Description

一种群智感知的多源信息融合室内定位方法及系统 技术领域
本发明涉及定位领域,具体而言,涉及一种群智感知的多源信息融合室内定位方法及系统。
背景技术
由于在商场、停车场、大型办公楼和医院的医疗保健等多个场景中室内定位技术的重要性越来越凸显出来。用于定位的全球定位系统(GPS)在室外环境中能提供良好的覆盖和高精度,但在室内环境中GPS信号被严重遮挡,因为卫星信号会被墙壁和天花板阻挡,因而其定位精度难以满足室内定位要求。所以在室内定位系统中应采用所有可用信息,例如WiFi、蓝牙、地磁场以及视觉图像信息等,而不是GPS来对目标位置进行估计。
传统的室内定位和跟踪方法是利用信号传播模型等处理无线传感器或物联网设备中的无线信号以导出目标位置,例如,到达时间(TOA)、到达时间差(TDOA)、到达角度(AOA)方法。但是,这些方法需要室内具备额外的设施以及对已部署设备的精确了解。因此利用指纹识别的方法更适用室内定位与跟踪,因为它不依赖于任何特定的信号传播模型。然而,利用单一的信号源信息如蓝牙、WiFi、地磁等对目标进行定位与跟踪很难满足定位的精度要求,因为这些信息源信号受到噪声干扰是不稳定的。
目前室内定位的解决方案可以划分为两大类:一是基于模型的室内定位算法,二是基于指纹库的定位算法。基于模型的室内定位算法由于受到模型本身的好坏以及信号测量所带来的误差的限制,而基于指纹库的室内定位算法则可以避免这一问题。基于指纹库的定位算法主要分为两个部分:一是离线采样阶段,一是在线定位部分。离线采样阶段主要是采集室内用于定位的传感器的信息,例如地磁强度的信息、蓝牙、WiFi的信号强度值以及图像等信息,在线阶段主要是采集目标所在位置的地磁、蓝牙、WiFi 以及图片等信息与之指纹库的信息进行匹配进而确定目标所在位置。但是在建立指纹库的过程中由于要确定多个采样点,所以要耗费极大的人力、物力和时间资源。此外室内WiFi、蓝牙等信息源的信号强度并不稳定,相对来说信号比较稳定的超带宽费用又较高。
因此,融合异构信息以增强位置估计是一种主要的有效性方法,即建立室内融合信息的指纹库进行定位。然而基于指纹信息融合的主要问题是需要收集大量数据来构建可靠的数据库,且建立这样的数据库消耗了大量的人力和时间。在这种情况下,人群感知(鼓励不同的移动用户共享其感知数据并上传到服务器)是数据库构建的有效方案。使用人群感知,移动用户可以从他们的智能手机共享与位置相关的不同信息,例如WiFi、地磁或视觉图像,使得定位系统可以基于这些信息导出准确的位置估计。一些视觉图像可以直接导致精确的位置,且地磁数据相当稳定,无论走路、跑步还是静止不动,WiFi数据也可以免费获取。
已有的室内定位算法有基于混合WiFi,地磁和航位推算(PDR)的智能手机室内导航方法,有基于指纹的室内定位的AP加权多重匹配最近邻法,有利用普遍存在的地磁和WiFi异常点进行定位的Magicol定位方法系统,有利用深度学习识别室内定位指纹库的定位方法,有使用地磁和图像传感器来定位的VMag系统,并对图像信息进行复杂的神经网络训练,还有只利用图像视觉进行定位的算法,该方法需要建立海量的图像指纹库信息并且与图像的每一帧特征相匹配因此耗时较多。
发明内容
本发明实施例提供了一种群智感知的多源信息融合室内定位方法及系统,以至少解决现有室内定位方法精确度低的技术问题。
根据本发明的一实施例,提供了一种群智感知的多源信息融合室内定位方法,包括以下步骤:
基于群体感知构建指纹识别数据库;
将采集到的地磁、WiFi和视觉图像信息进行融合并基于构建的指纹识别数据库对目标位置进行定位,其中在对目标位置进行定位中使用基于图像的小区域确定算法IBSM和分段结构自适应算法SSAC来估计和自适应地校准目标位置。
进一步地,基于群体感知构建指纹识别数据库包括:构建图像指纹识别并使用动态轨迹收集方法构建指纹识别数据库,指纹识别数据库包含地磁强度和WiFi的RSS值。
进一步地,基于群体感知构建指纹识别数据库包括:
特征区域的视觉图像的建立:构建图像数据库并将每个图像与相关位置进行链接;目标试图自我校准时搜索视觉信息并将其与图像数据库进行匹配;
WiFi和地磁动态指纹库的建立:采用动态轨迹收集来构建WiFi的RSS和地磁强度数据库,用户将智能手机沿着预定的无障碍路径在室内行走并上传数据,其中每个位置WiFi的RSS数据和地磁强度是同步获取的;在构建整个室内的数据收集之后,形成室内动态路径图。
进一步地,将采集到的地磁、WiFi和视觉图像信息进行融合并基于构建的指纹识别数据库对目标位置进行定位包括:
对典型的SIFT算法进行改进,使用圆形来构造SIFT特征描述符;
基于图像的子区域匹配法IBSM将每个采样图像对应于在创建图像指纹时确定的指定区域;
基于采用加权K近邻方法WKNN评估移动终端收集的每个WiFi无线接入点AP的RSS值与指纹中采样点对应的每个AP的RSS之间的相似度,确定具有高相似度的多个采样点的位置,并使用加权平均值推导出用户的位置;
对同步获取到的WiFi和地磁的指纹数据进行分段结构自适应校准, 在评估在线阶段获得的地磁序列与指纹识别中的地磁序列之间的相似性之后,确定两个序列的匹配。
进一步地,对典型的SIFT算法进行改进,使用圆形来构造SIFT特征描述符包括:
以关键点为中心,关键点周围的部分分为多个同心圆,从内到外分别有多种不同半径的同心圆,其半径分别为一个像素点、两个像素点直至多个像素点。
进一步地,对典型的SIFT算法进行改进,使用圆形来构造SIFT特征描述符还包括:使用随机样本共识方法RANSAC来消除不匹配点。
进一步地,基于图像的子区域匹配法IBSM将每个采样图像对应于在创建图像指纹时确定的指定区域包括:当目标接近某个采样点时,子区域匹配法IBSM中将直接给出一个特殊的位置坐标或子区域范围,然后通过WKNN算法测量精确的初始位置。
进一步地,对同步获取到的WiFi和地磁的指纹数据进行分段结构自适应校准包括:使用动态时间规整算法DTW将地磁强度序列视为连续波形,并在数据库中找到匹配点。
进一步地,对同步获取到的WiFi和地磁的指纹数据进行分段结构自适应校准还包括:地磁序列是分段的,校准后位置将更新,然后开始新的匹配校准。
根据本发明的另一实施例,提供了一种群智感知的多源信息融合室内定位系统,包括:
指纹识别数据库构建单元,用于基于群体感知构建指纹识别数据库;
定位单元,用于将采集到的地磁、WiFi和视觉图像信息进行融合并基于构建的指纹识别数据库对目标位置进行定位,其中在对目标位置进行定位中使用基于图像的小区域确定算法IBSM和分段结构自适应算法SSAC来 估计和自适应地校准目标位置。
本发明实施例中的群智感知的多源信息融合室内定位方法及系统,采用了一种基于群体感知的指纹识别技术,该技术效果显著,特别是对于复杂的基础设施。此外为了减少定位成本并提高定位精度,本发明收集了室内的三种免费资源信息即地磁、WiFi和视觉图像,由于单个信息源的信号强度是不稳定的,因此本发明将这三种信息源进行融合来对目标的位置进行估计。在系统中本发明还提出了两个新的算法:基于图像的小区域确定算法(IBSM)和分段结构自适应算法(SSAC),它们分别能够帮助目标快速确定所在室内的小区域范围或者确定目标所在的特殊位置以及帮助目标确定定位的准确性或者校准目标所在范围。
附图说明
此处所说明的附图用来提供对本发明的进一步理解,构成本申请的一部分,本发明的示意性实施例及其说明用于解释本发明,并不构成对本发明的不当限定。在附图中:
图1为本发明群智感知的多源信息融合室内定位方法的流程图;
图2为本发明中Wimage系统框架图;
图3为本发明中室内图像采样点分布图;
图4为本发明中描述符的矢量图;
图5为本发明中在相同路径上以不同步行速度的地磁数据曲线图;
图6为本发明中在相同路径上以不同手机姿态的地磁数据曲线图;
图7为本发明Wimage系统中IBSM和SSAC的性能比较图;
图8为本发明中WiFi信号强度图;
图9为本发明地磁校正在SSAC中的性能比较图;
图10为本发明中Wimage系统在室内的跟踪图效果图;
图11为本发明群智感知的多源信息融合室内定位系统的模块图。
具体实施方式
为了使本技术领域的人员更好地理解本发明方案,下面将结合本发明实施例中的附图,对本发明实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例仅仅是本发明一部分的实施例,而不是全部的实施例。基于本发明中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都应当属于本发明保护的范围。
需要说明的是,本发明的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象,而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换,以便这里描述的本发明的实施例能够以除了在这里图示或描述的那些以外的顺序实施。此外,术语“包括”和“具有”以及他们的任何变形,意图在于覆盖不排他的包含,例如,包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元,而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
本发明构建一个有效的框架Wimage,它结合了视觉图像、WiFi和地磁作为综合指纹数据库,根据它们的特性融合这些异构数据进而来推导目标位置。且图像视觉、地磁和WiFi信息在室内均是完全免费的,用户可以免费获取这些信息,并且不需要额外的设备布置。在Wimage内,视觉图像数据是一些参考图片,其指示与某些特定位置相关的一些关键特征,例如房间号、厨房和消防栓。这些采集的图像可以使用匹配算法尺度不变特征转换(SIFT)直接指示特殊位置或定位范围,有助于系统减少粗粒定位误差。同时通过人群感知来收集WiFi和地磁数据,并通过定位方法来定位目标位置,这些方法包含有加权k-最近邻(WKNN)和动态时间扭曲算法(DTW)。具体地,它使用基于WKNN算法的WiFi数据来估计目标位置,并且DTW算法的地磁数据用于通过分段地磁序列匹配进行校准和校正。在这种情况下,可以自动校正目标位置。该Wimage系统对一办公楼进行了 评估,对于多个实验下,均方根误差主要在0.5m以下。
此外定位的准确性和及时性也是室内定位的关键需求。本发明所提出的Wimage系统利用人群感知的方法对定位指纹库进行构建并且融合了图像视觉、地磁和WiFi三种信息设计定位算法。
1.基于指纹的室内定位算法在建立指纹库时为了保证定位的精度,需要在室内多个采样点大量采集相关数据,其中采样点是根据室内整体环境和条件对室内区域进行网格划分等方式来确定,网格点的稀疏或者稠密很大情况下影响了室内定位的精度,同时对未采样点也需要运用插值算法来确定指纹信息,这不仅会带来信息的不准确性也会耗费大量的人力、物力和时间成本,因为在建立指纹库时需要工作人员在室内划分好的采样点逐个去采集相关信息。这在前期建立指纹库的过程当中需要付出非常大的工作量。
为了解决以上问题,本发明在Wimage系统当中采用了一种人群感知的方法鼓励不同的移动用户共享其感知数据并上传到服务器,这种方法能够在建立指纹库的时候利用不同用户不同设备检测到的信息建立室内指纹库,同时本发明采用了一种不同于传统的离散点建库的方法,而是采用了一种路线轨迹的采集方式,工作人员可以拿着智能手机沿着室内的可行性路线同步采集地磁和WiFi信息,这极大的减少了指纹库的建立时间,因为工作人员只需要沿着室内路线轨迹行走便可采集到室内的地磁和WiFi信息,这种方法建立的指纹库也不需要借助于其他的插值算法来计算未采样点的指纹库信息。
2.由于室内受到建筑物的遮挡以及室内环境的复杂性,所以全球定位系统(GPS)无法运用在室内环境中,因而要借助于其他的定位方法来进行定位,目前室内能借助的有红外线、蓝牙、图像、WiFi、超带宽、地磁等,但是红外线、蓝牙、超带宽等都需要额外对室内进行部署,并且也增加了室内定位的开销。
为了解决上述问题,本发明的Wimage系统借助于室内的WiFi、地磁 以及图像三种资源进行定位。这三种室内资源用来定位不仅不需要对室内进行额外的部署而且是完全可行的。首先图像资源是随处都可获得的,且不同图片具有不同的特征,不同的图片特征可以针对不同的室内位置,这为利用图片进行定位提供了可能性。其次地球本身是一个巨大的磁场,因而地磁信息也是存在于室内各处的,而且不同的位置地磁场信息是不同的,所以可以利用不同的地磁信息来定位。最后,随着网络通信等的发展,WiFi在室内也越来越得到了普及,在办公楼、大型医院、大型会场基本都会有WiFi,且WiFi的信号强度随着与发射端口距离的远近其强度值是不同的,且某一位置会接收到不同的WiFi端口的不同WiFi信号强度(RSS)值,因此室内每个位置接收到的WiFi信息都是不同的,基于这种特征WiFi的RSS可以用来进行室内定位。
3.基于指纹的室内定位算法需要借助于某一定位技术如蓝牙、WiFi和红外线等,但是单一的信号源由于受到噪声或者障碍物等影响会存在信号波动不稳定的情况,因此使用单一的信号源进行定位很难精确的对目标位置进行估计。
本发明采用了图像、WiFi和地磁三种信息源进行定位,并且提出了可靠的定位算法,不仅能够快速的对目标进行初始位置的定位与跟踪还能够及时准确的对目标位置进行校正。具体做法是:利用室内特殊的图片用来具体确定目标所在位置或者范围,在这个过程中本发明采用了改进的SIFT算法进行图片的匹配,并且针对这整个过程设计提出了一种基于图片的小区域确定算法(IBSM),然后根据WiFi信号强度RSSI利用加权K最近邻算法对目标的初始位置进行估计,在对目标进行跟踪的同时利用地磁信息的轨迹变化特征对目标所在位置进行阶段性的校正,这里用到的方法为动态时间规整算法(DTW),这里的阶段性校正是因为在一段路径中地磁的变化是有规律的,而在建立地磁指纹库的过程中根据不同的路径也是阶段性的。针对目标位置的校正的过程本发明设计提出了一种分段结构自适应校准(SSAC)算法,实验表明它能够有效提高定位精度。
综上所述,为了解决以上问题并提高定位精度,本发明提出一种Wimage定位系统。本发明的Wimage定位系统采用了一种基于群体感知的指纹识别技术,该技术效果显著,特别是对于复杂的基础设施。此外为了减少定位成本并提高定位精度,本发明收集了室内的三种免费资源信息即地磁、WiFi和视觉图像,由于单个信息源的信号强度是不稳定的,因此本发明将这三种信息源进行融合来对目标的位置进行估计。在Wimage系统中本发明还提出了两个新的算法:基于图像的小区域确定算法(IBSM)和分段结构自适应算法(SSAC),它们分别能够帮助目标快速确定所在室内的小区域范围或者确定目标所在的特殊位置以及帮助目标确定定位的准确性或者校准目标所在范围。
本发明提供了利用室内免费资源进行位置估计的定位框架和算法的群智感知的多源信息融合室内定位方法及系统,该方法及系统中提出了使用人群感知的方法建立指纹库,在定位过程中提出了IBSM和SSAC算法,这能够有效减少定位成本以及人力物力资源的使用,并且能够有效提高室内定位精度,且具有较强的抗噪声抗干扰能力。
实施例1
根据本发明一实施例,提供了一种群智感知的多源信息融合室内定位方法,参见图1,包括以下步骤:
S101:基于群体感知构建指纹识别数据库;
S102:将采集到的地磁、WiFi和视觉图像信息进行融合并基于构建的指纹识别数据库对目标位置进行定位,其中在对目标位置进行定位中使用基于图像的小区域确定算法IBSM和分段结构自适应算法SSAC来估计和自适应地校准目标位置。
本发明实施例中的群智感知的多源信息融合室内定位方法,采用了一种基于群体感知的指纹识别技术,该技术效果显著,特别是对于复杂的基础设施。此外为了减少定位成本并提高定位精度,本发明收集了室内的三 种免费资源信息即地磁、WiFi和视觉图像,由于单个信息源的信号强度是不稳定的,因此本发明将这三种信息源进行融合来对目标的位置进行估计。在系统中本发明还提出了两个新的算法:基于图像的小区域确定算法(IBSM)和分段结构自适应算法(SSAC),它们分别能够帮助目标快速确定所在室内的小区域范围或者确定目标所在的特殊位置以及帮助目标确定定位的准确性或者校准目标所在范围。
具体地,本发明提出了一套完整的室内定位系统Wimage,它包括有两个阶段,即人群感知阶段和目标跟踪阶段。
Wimage系统的框架如图2所示,在人群感知阶段,本发明构建图像指纹识别并使用动态轨迹收集方法构建指纹识别数据库包含地磁强度和WiFi的RSS值。在目标跟踪阶段,目标基于收集的图像、地磁测量和WiFi信号执行多数据融合定位。在该部分中,应用多种方法,包括用于通过图像匹配确定子区域的改进SIFT算法、WKNN和DTW算法来估计和自适应地校准目标位置。具体包括:
1)基于人群感知的数据库建立
A.特征区域的视觉图像
在人群感知阶段,用户拍摄的图像与某些特定位置或子区域高度相关。在这种情况下,本发明可以将这些项目作为特征区域的语义信息。此外,如果目标试图自我校准,它可以使用相机搜索视觉信息并将其与图像数据库匹配。因此,本发明构建图像数据库并将每个图像与相关位置链接。如图3所示,本发明给出了一个室内图片的采集点及其所在位置,这些图片分别是消防栓、休息间以及门牌号等。
B.WiFi和地磁动态指纹库的建立
与基于采样点收集数据样本的收集方式不同,Wimage系统采用动态轨迹收集来构建WiFi的RSS和地磁强度数据库。用户只需要将智能手机沿着预定的无障碍路径在室内行走并上传他们的数据,这使得对用户来说更 加方便。另外,每个位置WiFi的RSS数据和地磁强度是同步获取的,因此它们的信息也是相互关联的。在构建整个室内的收集数据之后,形成室内动态路径图。
2)对目标的定位与追踪算法
A.典型的SIFT算法
基于图像的SIFT匹配算法是一种用于提取局部特征、搜索尺度空间中的极值点、空间尺度和旋转不变量的算法。SIFT算法适用于海量数据库中的快速准确匹配。图像匹配基于提取的SIFT特征,SIFT算法使用关键点来描述图像特征。关键点是浮动数字的向量,矢量的长度由描述符确定,描述符包含方向直方图阵列的方向和尺寸。
在典型的SIFT算法中,由4×4阵列和8个方向组成描述符的矢量,如图4所示,左图为典型SIFT算法的,右图为改进的SIFT算法的。假设R i=(r i1,r i2,...r i128)是匹配图像的SIFT描述符向量。并且S j=(s j1,s j2,...s j128)是作为提取N个关键点的被匹配图像的SIFT描述符矢量,并且j∈N。图像匹配实际上是根据欧几里德距离的测量来匹配的。假设本发明从R i和S m得到的min{Dis}为最近的欧几里德距离,R i和S t得到的SecMin{Dis}为次最近邻欧几里德距离,它们分别为:
Figure PCTCN2020128802-appb-000001
Figure PCTCN2020128802-appb-000002
通过最近邻距离与次近邻距离的比率的结果来查看它是否匹配。假设阈值T Dis,如果比率小于阈值,则匹配成功。如果比率大于阈值,则匹配不成功。SIFT算法中建议在一般情况下阈值T Dis为0.5。
Figure PCTCN2020128802-appb-000003
B.改进的SIFT算法
改进的SIFT算法主要是为了降低计算成本,提高匹配速度。在典型的SIFT算法中,描述符的维数较高,在Wimage系统中的图片匹配任务中,它将在低维的描述符中表现更好。在Wimage系统中,指纹识别中的图像通常有很大差异。因此,在改进的SIFT中,本发明考虑减小描述符的向量维度以提高匹配速度,使用圆形来构造SIFT特征描述符。以关键点为中心,关键点周围的部分分为6个同心圆,从内到外分别有三种不同半径的同心圆,其半径分别为1个像素点,两个像素点和三个像素点,在图8中用不同符号线表示。在典型的SIFT中,矢量维数为4×4×8=128,并在改进的SIFT中,矢量尺寸为6×8=48,这消耗较低的计算成本。匹配图像和被匹配图像的SIFT描述符矢量分别是R i=(r i1,r i2,...r i48)和S j=(s j1,s j2,...s j48)。同理根据公式(1)、(2)和(3)进行匹配。最后本发明用随机样本共识(RANSAC)来消除不匹配点,以确保匹配准确性。
C.基于图像的子区域匹配
在Wimage系统中,每个采样图像对应于在创建图像指纹时确定的指定区域。每个采样图像P将对应于坐标范围[X,Y],其中X表示[x,x']的范围,并且Y表示[y,y']的范围。图3显示了图片与子区域坐标之间的关系。图片匹配基于改进的SIFT特征,如果两个相似的图像P 0和P 1满足公式(3)阈值范围,那么目标的子区域将被快速锁定在与成功匹配的候选图像P 1相对应的范围[X 1,Y 1]中。本发明定义这种算法为基于图像的子区域匹配(IBSM)法。
在IBSM算法中,指纹识别中的图像包含与位置相关的详细语义信息。在图片匹配中,本发明使用改进的SIFT算法,该算法可行。目标可以使用图像匹配来确定定位子区域。当目标接近某个采样点时,它可以直接给出一个特殊的位置坐标或子区域范围,然后通过WKNN算法测量精确的初始位置。
D.WKNN算法
WiFi指纹数据库的匹配算法采用加权K近邻(WKNN)方法,该方法评估移动终端收集的每个WiFi无线接入点(AP)的RSS值与指纹中采样点对应的每个AP的RSS之间的相似度。(相似度的衡量一般采用欧氏距离)确定具有高相似度的k个采样点的位置,并使用加权平均值推导出用户的位置:
Figure PCTCN2020128802-appb-000004
其中(x i,y i)是对应于第i个相邻参考点的坐标,(x,y)是估计的位置,并且w i是第i个相邻点的权重。
对于动态指纹库的构建,本发明在构建轨迹数据库时进行多次采样,每个轨迹的指纹组成可写为:D=[D 1,D 2,...D h],其中h是轨迹指纹的样本大小。每个轨迹指纹数据库的组成是:
Figure PCTCN2020128802-appb-000005
其中t是第h个采样轨道的长度。假设在第h次采样的轨迹中采样点j处收集的AP信号强度为
Figure PCTCN2020128802-appb-000006
其中
Figure PCTCN2020128802-appb-000007
是在第h次采样的轨迹处采样点j处收集的第n个AP的RSS值。然后,对于每次采样,在相同采样点相同的WiFi信号强度值会出现不一样的值,因此本发明可以计算它的方差值σ j=(σ j1j2,...σ jn)。方差反映了样本数据分布的分散。方差越大,rss值的波动越大,所以权重的表达式为:
Figure PCTCN2020128802-appb-000008
E.分段结构自适应校准
为了提高定位精度,本发明采用地磁校准。在Wimage系统中,同步获取WiFi和地磁的指纹数据,为地磁校准提供了可行性。本发明采用动态轨迹收集来构建地磁指纹库。因此,地磁数据可以被认为是多个序列段。 且本发明知道不同用户的步行速度不同,在相同路径收集的地磁波形状是相似的,如图5和图6所示。而动态时间规整算法(DTW)是一种测量不同长度的两个时间序列相似度的方法。它可以动态匹配两个相似波形的点。在目标跟踪阶段,DTW将地磁强度序列视为连续波形,并在数据库中找到匹配点。假设指纹库中的地磁路径序列是:Z={z 1,z 2,...z m},需要匹配的路径序列是:X={x 1,x 2,...x n}。Z和X需要延伸或缩短以在形状上对齐。规划路径W表示这种映射关系:
W={w 1,w 2,w 3...w k}  (6)
其中max{m,n}≤k≤m+n,并且w 1=(z 1,x 1),w k=(z m,x n)。最低成本路径是:
Figure PCTCN2020128802-appb-000009
根据动态规划,最小成本路径也可以表示:
D(Z,X)=dist(z i,x j)+min[D(z i-1,x j),D(z i,x j-1),D(z i-1,x j-1)]  (8)
它是两个路径序列相似性的测量标准。在评估在线阶段获得的地磁序列与指纹识别中的地磁序列之间的相似性之后,可以确定两个序列的匹配。
地磁校准对于定位的准确性具有重要意义。它可以检查WiFi定位的结果,确认定位结果的准确性或纠正错误的定位。考虑到此功能,本发明在细分上匹配地磁数据。此外,虽然使用单个地磁数据进行定位会受到噪声的影响,但对于序列段匹配结果更可靠,定义这种方法称为分段结构自适应校准(SSAC)的算法。此外,地磁序列是分段的,校准后位置将更新,然后开始新的匹配校准,这也可以避免累积误差。
实施例2
根据本发明的另一实施例,提供了一种群智感知的多源信息融合室内定位系统,参见图11,包括:
指纹识别数据库构建单元100,用于基于群体感知构建指纹识别数据 库;
定位单元200,用于将采集到的地磁、WiFi和视觉图像信息进行融合并基于构建的指纹识别数据库对目标位置进行定位,其中在对目标位置进行定位中使用基于图像的小区域确定算法IBSM和分段结构自适应算法SSAC来估计和自适应地校准目标位置。
本发明实施例中的群智感知的多源信息融合室内定位系统,采用了一种基于群体感知的指纹识别技术,该技术效果显著,特别是对于复杂的基础设施。此外为了减少定位成本并提高定位精度,本发明收集了室内的三种免费资源信息即地磁、WiFi和视觉图像,由于单个信息源的信号强度是不稳定的,因此本发明将这三种信息源进行融合来对目标的位置进行估计。在系统中本发明还提出了两个新的算法:基于图像的小区域确定算法(IBSM)和分段结构自适应算法(SSAC),它们分别能够帮助目标快速确定所在室内的小区域范围或者确定目标所在的特殊位置以及帮助目标确定定位的准确性或者校准目标所在范围。
具体地,本发明提出了一套完整的室内定位系统Wimage,它包括有两个阶段,即人群感知阶段和目标跟踪阶段。
Wimage系统的框架如图2所示,在人群感知阶段,本发明构建图像指纹识别并使用动态轨迹收集方法构建指纹识别数据库包含地磁强度和WiFi的RSS值。在目标跟踪阶段,目标基于收集的图像、地磁测量和WiFi信号执行多数据融合定位。在该部分中,应用多种方法,包括用于通过图像匹配确定子区域的改进SIFT算法、WKNN和DTW算法来估计和自适应地校准目标位置。具体包括:
1)指纹识别数据库构建单元100:基于人群感知的数据库建立
A.特征区域的视觉图像
在人群感知阶段,用户拍摄的图像与某些特定位置或子区域高度相关。在这种情况下,本发明可以将这些项目作为特征区域的语义信息。此外, 如果目标试图自我校准,它可以使用相机搜索视觉信息并将其与图像数据库匹配。因此,本发明构建图像数据库并将每个图像与相关位置链接。如图3所示,本发明给出了一个室内图片的采集点及其所在位置,这些图片分别是消防栓、休息间以及门牌号等。
B.WiFi和地磁动态指纹库的建立
与基于采样点收集数据样本的收集方式不同,Wimage系统采用动态轨迹收集来构建WiFi的RSS和地磁强度数据库。用户只需要将智能手机沿着预定的无障碍路径在室内行走并上传他们的数据,这使得对用户来说更加方便。另外,每个位置WiFi的RSS数据和地磁强度是同步获取的,因此它们的信息也是相互关联的。在构建整个室内的收集数据之后,形成室内动态路径图。
2)定位单元200:对目标的定位与追踪算法
A.典型的SIFT算法
基于图像的SIFT匹配算法是一种用于提取局部特征、搜索尺度空间中的极值点、空间尺度和旋转不变量的算法。SIFT算法适用于海量数据库中的快速准确匹配。图像匹配基于提取的SIFT特征,SIFT算法使用关键点来描述图像特征。关键点是浮动数字的向量,矢量的长度由描述符确定,描述符包含方向直方图阵列的方向和尺寸。
在典型的SIFT算法中,由4×4阵列和8个方向组成描述符的矢量,如图4所示,左图为典型SIFT算法的,右图为改进的SIFT算法的。假设R i=(r i1,r i2,...r i128)是匹配图像的SIFT描述符向量。并且S j=(s j1,s j2,...s j128)是作为提取N个关键点的被匹配图像的SIFT描述符矢量,并且j∈N。图像匹配实际上是根据欧几里德距离的测量来匹配的。假设本发明从R i和S m得到的min{Dis}为最近的欧几里德距离,R i和S t得到的SecMin{Dis}为次最近邻欧几里德距离,它们分别为:
Figure PCTCN2020128802-appb-000010
Figure PCTCN2020128802-appb-000011
通过最近邻距离与次近邻距离的比率的结果来查看它是否匹配。假设阈值T Dis,如果比率小于阈值,则匹配成功。如果比率大于阈值,则匹配不成功。SIFT算法中建议在一般情况下阈值T Dis为0.5。
Figure PCTCN2020128802-appb-000012
B.改进的SIFT算法
改进的SIFT算法主要是为了降低计算成本,提高匹配速度。在典型的SIFT算法中,描述符的维数较高,在Wimage系统中的图片匹配任务中,它将在低维的描述符中表现更好。在Wimage系统中,指纹识别中的图像通常有很大差异。因此,在改进的SIFT中,本发明考虑减小描述符的向量维度以提高匹配速度,使用圆形来构造SIFT特征描述符。以关键点为中心,关键点周围的部分分为6个同心圆,从内到外分别有三种不同半径的同心圆,其半径分别为1个像素点,两个像素点和三个像素点,在图8中用不同符号线表示。在典型的SIFT中,矢量维数为4×4×8=128,并在改进的SIFT中,矢量尺寸为6×8=48,这消耗较低的计算成本。匹配图像和被匹配图像的SIFT描述符矢量分别是R i=(r i1,r i2,...r i48)和S j=(s j1,s j2,...s j48)。同理根据公式(1)、(2)和(3)进行匹配。最后本发明用随机样本共识(RANSAC)来消除不匹配点,以确保匹配准确性。
C.基于图像的子区域匹配
在Wimage系统中,每个采样图像对应于在创建图像指纹时确定的指定区域。每个采样图像P将对应于坐标范围[X,Y],其中X表示[x,x']的范围,并且Y表示[y,y']的范围。图3显示了图片与子区域坐标之间的关系。图片匹配基于改进的SIFT特征,如果两个相似的图像P 0和P 1满足公式(3)阈值范围,那么目标的子区域将被快速锁定在与成功匹配的候选图像P 1相对应的范围[X 1,Y 1]中。本发明定义这种算法为基于图像的子区域匹配(IBSM) 法。
在IBSM算法中,指纹识别中的图像包含与位置相关的详细语义信息。在图片匹配中,本发明使用改进的SIFT算法,该算法可行。目标可以使用图像匹配来确定定位子区域。当目标接近某个采样点时,它可以直接给出一个特殊的位置坐标或子区域范围,然后通过WKNN算法测量精确的初始位置。
D.WKNN算法
WiFi指纹数据库的匹配算法采用加权K近邻(WKNN)方法,该方法评估移动终端收集的每个WiFi无线接入点(AP)的RSS值与指纹中采样点对应的每个AP的RSS之间的相似度。(相似度的衡量一般采用欧氏距离)确定具有高相似度的k个采样点的位置,并使用加权平均值推导出用户的位置:
Figure PCTCN2020128802-appb-000013
其中(x i,y i)是对应于第i个相邻参考点的坐标,(x,y)是估计的位置,并且w i是第i个相邻点的权重。
对于动态指纹库的构建,本发明在构建轨迹数据库时进行多次采样,每个轨迹的指纹组成可写为:D=[D 1,D 2,...D h],其中h是轨迹指纹的样本大小。每个轨迹指纹数据库的组成是:
Figure PCTCN2020128802-appb-000014
其中t是第h个采样轨道的长度。假设在第h次采样的轨迹中采样点j处收集的AP信号强度为
Figure PCTCN2020128802-appb-000015
其中
Figure PCTCN2020128802-appb-000016
是在第h次采样的轨迹处采样点j处收集的第n个AP的RSS值。然后,对于每次采样,在相同采样点相同的WiFi信号强度值会出现不一样的值,因此本发明可以计算它的方差值σ j=(σ j1j2,...σ jn)。方差反映了样本数据分布的分散。方差越大,rss值的波动越大,所以权重的表达式为:
Figure PCTCN2020128802-appb-000017
E.分段结构自适应校准
为了提高定位精度,本发明采用地磁校准。在Wimage系统中,同步获取WiFi和地磁的指纹数据,为地磁校准提供了可行性。本发明采用动态轨迹收集来构建地磁指纹库。因此,地磁数据可以被认为是多个序列段。且本发明知道不同用户的步行速度不同,在相同路径收集的地磁波形状是相似的,如图5和图6所示。而动态时间规整算法(DTW)是一种测量不同长度的两个时间序列相似度的方法。它可以动态匹配两个相似波形的点。在目标跟踪阶段,DTW将地磁强度序列视为连续波形,并在数据库中找到匹配点。假设指纹库中的地磁路径序列是:Z={z 1,z 2,...z m},需要匹配的路径序列是:X={x 1,x 2,...x n}。Z和X需要延伸或缩短以在形状上对齐。规划路径W表示这种映射关系:
W={w 1,w 2,w 3...w k}  (6)
其中max{m,n}≤k≤m+n,并且w 1=(z 1,x 1),w k=(z m,x n)。最低成本路径是:
Figure PCTCN2020128802-appb-000018
根据动态规划,最小成本路径也可以表示:
D(Z,X)=dist(z i,x j)+min[D(z i-1,x j),D(z i,x j-1),D(z i-1,x j-1)]  (8)
它是两个路径序列相似性的测量标准。在评估在线阶段获得的地磁序列与指纹识别中的地磁序列之间的相似性之后,可以确定两个序列的匹配。
地磁校准对于定位的准确性具有重要意义。它可以检查WiFi定位的结果,确认定位结果的准确性或纠正错误的定位。考虑到此功能,本发明在细分上匹配地磁数据。此外,虽然使用单个地磁数据进行定位会受到噪 声的影响,但对于序列段匹配结果更可靠,定义这种方法称为分段结构自适应校准(SSAC)的算法。此外,地磁序列是分段的,校准后位置将更新,然后开始新的匹配校准,这也可以避免累积误差。
本发明的创新点至少在于:
1)在本发明中提出的完整的室内定位系统Wimage;
2)在本发明中提出的利用动态轨迹同步采集地磁和WiFi信息的方法;
3)本发明中提出的基于图像的小区域定位算法(IBSM);
4)在图像匹配中提出的改进的SIFT算法;
5)利用地磁信息进行的分段结构自适应校准算法(SSAM)。
本发明的优点至少在于:
优点一:在图像匹配时,本发明提出了改进的SIFT算法,改进的SIFT算法简化了图像的特征描述符,同时保证了图像匹配在Wimage系统中匹配的准确性,简化的特征描述符极大的降低了计算资源和成本,同时提高了图像匹配的效率;
优点二:本发明在定位时选取的多源信息分别是WiFi、地磁和图像,这三种信息在室内均是免费可以利用的;
优点三:在定位的离线建库阶段,本发明采用了人群感知的方法来节约工作人员建立指纹库的时间和物力资源,并且在采集指纹库的时候,本发明提出了利用轨迹规划的方式同步采集WiFi和地磁信息,这种方法只需沿着室内的路径行走便可获得室内WiFi信号强度和地磁的强度值,简化了建立指纹库的构建方式并且提高了指纹库的精确性;
优点四:本发明提出了利用地磁进行分段结构的自适应校准方案,这能够及时的为目标校正位置,并且极大的提高了定位精度。
本发明的实际实践如下:
A.实验设置
本发明提出的Wimage在综合实验中进行评估。构建了包含室内地标,RSS数据库和地磁强度数据库和可视化图像数据库。在人群感知阶段,所有RSS和地磁强度数据都是通过动态轨迹收集来收集的。本发明实验中的AP总数为26,WiFi的扫描周期设置为0.02s。
B.IBSM评估
IBSM在Wimage中的作用主要是减少初始定位中的定位范围并准确地将目标划分为小区域。确定采样点的位置或子区域主要是使用图像匹配。本发明采用改进的SIFT,在满足匹配精度时大大降低了计算成本。如图7所示,本发明可以从累计误差看出IBSM可以帮助目标提高定位精度。
C.SSAC评估
SSAC是Wimage的重要组成部分。WKNN算法使用WiFi数据来定位目标,但是由于环境影响,WiFi数据是非常不稳定的。本发明在一条路径上随机跟踪了三个WiFi端口的信号强度,图8反映了WiFi强度的波动。因此,定位的准确性将受到很大影响。本发明评估了地磁校准的性能,如图9所示,在SSAC算法中,地磁校准提高了定位精度。本发明使用Wimage系统跟踪办公楼中的一条路线,如图10所示,实线条代表实际路线,点虚线条代表估算路线。可以清楚地看到估算路线接近实际路线。另外,本发明在Wimage中测量IBSM和SSAC算法。本发明可以在图7中看到,对目标位置的准确性有很大影响。此外,校正地磁数据非常方便,因为地磁无处不在。
在本发明中的改进SIFT算法也可以直接利用SIFT算法进行图片的匹配,只是会增加计算资源和时间;本发明设计的方案不仅能够在办公楼中应用同时也能够在医院、大型会场以及商场等场景中应用。
上述本发明实施例序号仅仅为了描述,不代表实施例的优劣。
在本发明的上述实施例中,对各个实施例的描述都各有侧重,某个实 施例中没有详述的部分,可以参见其他实施例的相关描述。
在本申请所提供的几个实施例中,应该理解到,所揭露的技术内容,可通过其它的方式实现。其中,以上所描述的系统实施例仅仅是示意性的,例如单元的划分,可以为一种逻辑功能划分,实际实现时可以有另外的划分方式,例如多个单元或组件可以结合或者可以集成到另一个系统,或一些特征可以忽略,或不执行。另一点,所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口,单元或模块的间接耦合或通信连接,可以是电性或其它的形式。
作为分离部件说明的单元可以是或者也可以不是物理上分开的,作为单元显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外,在本发明各个实施例中的各功能单元可以集成在一个处理单元中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现,也可以采用软件功能单元的形式实现。
集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时,可以存储在一个计算机可读取存储介质中。基于这样的理解,本发明的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质中,包括若干指令用以使得一台计算机设备(可为个人计算机、服务器或者网络设备等)执行本发明各个实施例方法的全部或部分步骤。而前述的存储介质包括:U盘、只读存储器(ROM,Read-Only Memory)、随机存取存储器(RAM,Random Access Memory)、移动硬盘、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述仅是本发明的优选实施方式,应当指出,对于本技术领域的普通技术人员来说,在不脱离本发明原理的前提下,还可以做出若干改进 和润饰,这些改进和润饰也应视为本发明的保护范围。

Claims (10)

  1. 一种群智感知的多源信息融合室内定位方法,其特征在于,包括以下步骤:
    基于群体感知构建指纹识别数据库;
    将采集到的地磁、WiFi和视觉图像信息进行融合并基于构建的所述指纹识别数据库对目标位置进行定位,其中在对目标位置进行定位中使用基于图像的小区域确定算法IBSM和分段结构自适应算法SSAC来估计和自适应地校准目标位置。
  2. 根据权利要求1所述的群智感知的多源信息融合室内定位方法,其特征在于,所述基于群体感知构建指纹识别数据库包括:构建图像指纹识别并使用动态轨迹收集方法构建指纹识别数据库,所述指纹识别数据库包含地磁强度和WiFi的RSS值。
  3. 根据权利要求2所述的群智感知的多源信息融合室内定位方法,其特征在于,所述基于群体感知构建指纹识别数据库包括:
    特征区域的视觉图像的建立:构建图像数据库并将每个图像与相关位置进行链接;目标试图自我校准时搜索视觉信息并将其与图像数据库进行匹配;
    WiFi和地磁动态指纹库的建立:采用动态轨迹收集来构建WiFi的RSS和地磁强度数据库,用户将智能手机沿着预定的无障碍路径在室内行走并上传数据,其中每个位置WiFi的RSS数据和地磁强度是同步获取的;在构建整个室内的数据收集之后,形成室内动态路径图。
  4. 根据权利要求1所述的群智感知的多源信息融合室内定位方法,其特征在于,所述将采集到的地磁、WiFi和视觉图像信息进行融合并基于构建的所述指纹识别数据库对目标位置进行定位包括:
    对典型的SIFT算法进行改进,使用圆形来构造SIFT特征描述符;
    基于图像的子区域匹配法IBSM将每个采样图像对应于在创建图像指纹时确定的指定区域;
    基于采用加权K近邻方法WKNN评估移动终端收集的每个WiFi无线接入点AP的RSS值与指纹中采样点对应的每个AP的RSS之间的相似度,确定具有高相似度的多个采样点的位置,并使用加权平均值推导出用户的位置;
    对同步获取到的WiFi和地磁的指纹数据进行分段结构自适应校准,在评估在线阶段获得的地磁序列与指纹识别中的地磁序列之间的相似性之后,确定两个序列的匹配。
  5. 根据权利要求4所述的群智感知的多源信息融合室内定位方法,其特征在于,所述对典型的SIFT算法进行改进,使用圆形来构造SIFT特征描述符包括:
    以关键点为中心,关键点周围的部分分为多个同心圆,从内到外分别有多种不同半径的同心圆,其半径分别为一个像素点、两个像素点直至多个像素点。
  6. 根据权利要求5所述的群智感知的多源信息融合室内定位方法,其特征在于,所述对典型的SIFT算法进行改进,使用圆形来构造SIFT特征描述符还包括:使用随机样本共识方法RANSAC来消除不匹配点。
  7. 根据权利要求4所述的群智感知的多源信息融合室内定位方法,其特征在于,所述基于图像的子区域匹配法IBSM将每个采样图像对应于在创建图像指纹时确定的指定区域包括:当目标接近某个采样点时,子区域匹配法IBSM中将直接给出一个特殊的位置坐标或子区域范围,然后通过WKNN算法测量精确的初始位置。
  8. 根据权利要求4所述的群智感知的多源信息融合室内定位方法,其特征在于,所述对同步获取到的WiFi和地磁的指纹数据进行分段结构自适应校准包括:使用动态时间规整算法DTW将地磁强度序列视为连续波形, 并在数据库中找到匹配点。
  9. 根据权利要求8所述的群智感知的多源信息融合室内定位方法,其特征在于,所述对同步获取到的WiFi和地磁的指纹数据进行分段结构自适应校准还包括:地磁序列是分段的,校准后位置将更新,然后开始新的匹配校准。
  10. 一种群智感知的多源信息融合室内定位系统,其特征在于,包括:
    指纹识别数据库构建单元,用于基于群体感知构建指纹识别数据库;
    定位单元,用于将采集到的地磁、WiFi和视觉图像信息进行融合并基于构建的所述指纹识别数据库对目标位置进行定位,其中在对目标位置进行定位中使用基于图像的小区域确定算法IBSM和分段结构自适应算法SSAC来估计和自适应地校准目标位置。
PCT/CN2020/128802 2019-11-14 2020-11-13 一种群智感知的多源信息融合室内定位方法及系统 WO2021093872A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911113600.8 2019-11-14
CN201911113600.8A CN110856112B (zh) 2019-11-14 2019-11-14 一种群智感知的多源信息融合室内定位方法及系统

Publications (1)

Publication Number Publication Date
WO2021093872A1 true WO2021093872A1 (zh) 2021-05-20

Family

ID=69601795

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/128802 WO2021093872A1 (zh) 2019-11-14 2020-11-13 一种群智感知的多源信息融合室内定位方法及系统

Country Status (2)

Country Link
CN (1) CN110856112B (zh)
WO (1) WO2021093872A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114742995A (zh) * 2022-05-13 2022-07-12 盈嘉互联(北京)科技有限公司 一种基于数字孪生建筑和异质特征融合的室内定位方法
CN114845388A (zh) * 2022-05-17 2022-08-02 电子科技大学 一种分方向熵加权wknn的位置指纹室内定位方法

Families Citing this family (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110856112B (zh) * 2019-11-14 2021-06-18 深圳先进技术研究院 一种群智感知的多源信息融合室内定位方法及系统
CN111427982A (zh) * 2020-03-20 2020-07-17 浙江大学城市学院 一种基于群智感知的WiFi室内定位指纹数据库构建及更新技术
CN111405469B (zh) * 2020-03-24 2021-06-01 辽宁大学 基于手机移动传感网络的矿震监测系统及群智定位方法
CN111491367B (zh) * 2020-04-20 2021-03-30 电子科技大学 一种基于群智感知和多融合技术的室内定位方法
CN111521971B (zh) * 2020-05-13 2021-04-09 北京洛必德科技有限公司 一种机器人的定位方法及系统
CN111935644B (zh) * 2020-08-10 2021-08-24 腾讯科技(深圳)有限公司 一种基于融合信息的定位方法、装置及终端设备
CN112013842A (zh) * 2020-08-29 2020-12-01 桂林电子科技大学 一种基于图像地磁场和惯性传感器的多模态室内定位方法
CN112165684B (zh) * 2020-09-28 2021-09-14 上海大学 基于联合视觉和无线信号特征的高精度室内定位方法
CN112325883B (zh) * 2020-10-19 2022-06-21 湖南大学 一种WiFi和视觉多源融合的移动机器人室内定位方法
CN113111778B (zh) * 2021-04-12 2022-11-15 内蒙古大学 一种视频和无线融合的大规模人群分析方法
CN113074718B (zh) * 2021-04-27 2024-03-29 广东电网有限责任公司清远供电局 一种定位方法、装置、设备及存储介质
CN113490146B (zh) * 2021-05-08 2022-04-22 湖南大学 一种基于WiFi与视觉融合的SLAM方法
CN113382376B (zh) * 2021-05-08 2022-05-10 湖南大学 一种基于wifi和视觉集成的室内定位方法
CN113490137B (zh) * 2021-05-08 2022-05-03 湖南大学 一种基于WiFi与视觉融合室内定位方法
CN113259883B (zh) * 2021-05-18 2023-01-31 南京邮电大学 一种面向手机用户的多源信息融合的室内定位方法
CN113347563B (zh) * 2021-05-31 2021-12-07 广东技术师范大学 一种基于移动群智感知任务模型的指纹库构建方法及设备
CN114612731B (zh) * 2022-05-10 2022-08-09 安徽省路通公路工程检测有限公司 一种用于道路平整度检测的智能识别方法及系统
CN116311396B (zh) * 2022-08-18 2023-12-12 荣耀终端有限公司 用于指纹识别的方法和装置
CN116184312B (zh) * 2022-12-22 2023-11-21 泰州雷德波达定位导航科技有限公司 一种基于语义Wi-Fi的室内众源指纹库构建方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105792353A (zh) * 2016-03-14 2016-07-20 中国人民解放军国防科学技术大学 群智感知式WiFi信号指纹辅助的图像匹配室内定位方法
WO2016122019A1 (ko) * 2015-01-28 2016-08-04 (주)휴빌론 Wps(wifi positioning system)를 이용한 이동 방향 추정 방법 및 시스템
CN108668245A (zh) * 2018-04-13 2018-10-16 哈尔滨工业大学 一种基于WiFi与地磁场数据库的室内定位方法
CN109945865A (zh) * 2019-02-25 2019-06-28 天津大学 基于WiFi与地磁融合的室内定位方法
CN110856112A (zh) * 2019-11-14 2020-02-28 深圳先进技术研究院 一种群智感知的多源信息融合室内定位方法及系统

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102645208B (zh) * 2012-04-01 2014-05-14 燕山大学 基于动态路由机制的视觉测量定位及校正系统
CN105160686B (zh) * 2015-10-21 2017-08-25 武汉大学 一种基于改进sift算子的低空多视角遥感影像匹配方法
CN105716604A (zh) * 2016-02-25 2016-06-29 华南理工大学 基于地磁序列的移动机器人室内定位方法及系统
CN107339990B (zh) * 2017-06-27 2020-05-08 北京邮电大学 多模式融合定位系统及方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2016122019A1 (ko) * 2015-01-28 2016-08-04 (주)휴빌론 Wps(wifi positioning system)를 이용한 이동 방향 추정 방법 및 시스템
CN105792353A (zh) * 2016-03-14 2016-07-20 中国人民解放军国防科学技术大学 群智感知式WiFi信号指纹辅助的图像匹配室内定位方法
CN108668245A (zh) * 2018-04-13 2018-10-16 哈尔滨工业大学 一种基于WiFi与地磁场数据库的室内定位方法
CN109945865A (zh) * 2019-02-25 2019-06-28 天津大学 基于WiFi与地磁融合的室内定位方法
CN110856112A (zh) * 2019-11-14 2020-02-28 深圳先进技术研究院 一种群智感知的多源信息融合室内定位方法及系统

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114742995A (zh) * 2022-05-13 2022-07-12 盈嘉互联(北京)科技有限公司 一种基于数字孪生建筑和异质特征融合的室内定位方法
CN114742995B (zh) * 2022-05-13 2022-09-20 盈嘉互联(北京)科技有限公司 一种基于数字孪生建筑和异质特征融合的室内定位方法
CN114845388A (zh) * 2022-05-17 2022-08-02 电子科技大学 一种分方向熵加权wknn的位置指纹室内定位方法
CN114845388B (zh) * 2022-05-17 2023-02-28 电子科技大学 一种分方向熵加权wknn的位置指纹室内定位方法

Also Published As

Publication number Publication date
CN110856112A (zh) 2020-02-28
CN110856112B (zh) 2021-06-18

Similar Documents

Publication Publication Date Title
WO2021093872A1 (zh) 一种群智感知的多源信息融合室内定位方法及系统
Vo et al. A survey of fingerprint-based outdoor localization
Huang et al. WiFi and vision-integrated fingerprint for smartphone-based self-localization in public indoor scenes
JP7247186B2 (ja) 室内位置決め方法、室内位置決めシステム、室内位置決めデバイスおよびコンピュータ読み取り可能媒体
WO2019062734A1 (zh) 基于Wi-Fi热点的室内定位方法及装置
JP2015531053A (ja) 無線マップを動的に作成するためのシステム、方法、及びコンピュータプログラム
Wu et al. EasiTrack: Decimeter-level indoor tracking with graph-based particle filtering
Zhang et al. A scalable indoor localization algorithm based on distance fitting and fingerprint mapping in Wi-Fi environments
Feng et al. Visual Map Construction Using RGB‐D Sensors for Image‐Based Localization in Indoor Environments
Kim et al. Crowdsensing-based Wi-Fi radio map management using a lightweight site survey
Jang et al. Survey of landmark-based indoor positioning technologies
He et al. WiFi iLocate: WiFi based indoor localization for smartphone
Zhou et al. Anonymous crowdsourcing-based WLAN indoor localization
Gong et al. A usability-enhanced smartphone indoor positioning solution using compressive sensing
Yang et al. Deepwipos: a deep learning-based wireless positioning framework to address fingerprint instability
Shu et al. 3d point cloud-based indoor mobile robot in 6-dof pose localization using a wi-fi-aided localization system
Chen et al. Path distance-based map matching for Wi-Fi fingerprinting positioning
Zhou et al. IMLours: Indoor mapping and localization using time-stamped WLAN received signal strength
Jia et al. A fingerprint-based localization algorithm based on LSTM and data expansion method for sparse samples
GB2570853A (en) Identifying sites visited by a user device
CN111836188A (zh) 一种基于Wi-Fi RSS的在线协同定位及系统
Wu et al. Indoor surveillance video based feature recognition for pedestrian dead reckoning
He et al. Vision-aided self-calibration of a wireless propagation model for crowdsourcing-based indoor localization
CN114679683A (zh) 一种基于派生指纹迁移的室内智能定位方法
He et al. Voronoi diagram based indoor localization in wireless sensor networks

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20888487

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20888487

Country of ref document: EP

Kind code of ref document: A1