Accurate positioning method and system based on ultrahigh frequency RFID electronic tag

Through wavelet transformation fusion technology and the position prediction model embedded in depth information features, combined with visual information, the position robustness and accuracy of RFID tags in complex environments are solved, and higher positioning accuracy is achieved.

CN119066336APending Publication Date: 2024-12-03刘妃
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411094646.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-10
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The existing RFID tags have poor positioning robustness in complex environments and have failed to effectively use visual information to improve positioning accuracy.

Method used

The fingerprint data is preprocessed by wavelet transform fusion technology, combined with the position prediction model embedded in the depth information feature, and optimize the position prediction model through mixed positioning errors, and use visual information and depth information to improve positioning accuracy.

Benefits of technology

It improves the positioning robustness and accuracy of RFID tags in complex environments, enhances the model's resistance to multipath effects and interference, and improves the accuracy of position prediction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119066336A_ABST
    Figure CN119066336A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of electronic tags, in particular to a precise positioning method and system based on an ultrahigh frequency RFID electronic tag. Firstly, fingerprint data of all reference mobile tags are obtained; secondly, the collected fingerprint data is preprocessed, and the first fingerprint data after mean filtering and the second fingerprint data after residual Gaussian filtering are fused by using a wavelet transform fusion technology, so that the robustness of the model in a complex signal environment is effectively improved; then, a position prediction model based on depth information feature embedding is provided, and the model can accurately obtain the position information of the mobile tag by using a depth information guiding scheme; and finally, a hybrid positioning error is proposed for training the position prediction model, and the error calculates a fingerprint feature difference value and a predicted position difference value through weighting, so that the prediction accuracy of the position prediction model is effectively improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of electronic tags, and in particular to a precise positioning method and system based on ultra-high frequency RFID electronic tags. Background Art

[0002] The current global satellite navigation system can provide people with ready location information in outdoor environments, greatly facilitating people's travel. However, in indoor environments, the non-line-of-sight propagation and multipath effects caused by building occlusion and the complexity of the indoor environment have greatly limited the personnel positioning technology based on satellite navigation.

[0003] At present, RFID technology has been widely used in indoor positioning. When accurately locating RFID tags, it mainly relies on the nonlinear mapping model between signal strength (RSSI) and distance. The basic principle of RSSI positioning technology is that the attenuation of the radio frequency signal is inversely proportional to the square of the distance, and the signal transmission distance is inferred by measuring the power strength of the received signal. However, due to possible interference in the environment, this mapping relationship will fluctuate. Therefore, studying how to quantify the relationship between RSSI and distance has become one of the key issues in solving RFID tag positioning technology.

[0004] With the continuous development of technology, deep learning has become an emerging technology in the field of big data processing and an effective tool for learning the relationship between RSSI and distance. However, there are still some defects in the existing technical literature for indoor positioning based on RFID tags. For example, the Chinese patent with patent publication number CN111537952B proposes a RFID indoor three-dimensional positioning solution based on deep learning. The solution combines relative positioning and absolute positioning, uses deep learning technology, further mines data features, and achieves better RFID three-dimensional indoor positioning effect. The positioning model is stable and reliable and can be applied in real life scenarios. For example, the Chinese patent with patent publication number CN114710831B proposes an RFID tag positioning system based on deep learning. The system trains the convolutional neural network by constructing an RSS / PDOA joint fingerprint database, so that it has a good classification effect to achieve higher indoor positioning accuracy. For example, the Chinese patent with patent publication number CN115103299B proposes a multi-sensor fusion positioning method based on RFID. The method uses a deep neural network to establish a nonlinear relationship between fingerprint features and position coordinates, and uses a Kalman filter to fuse data collected by multiple sensors, thereby improving positioning accuracy.

[0005] The above existing technologies perform a single filtering operation on the input fingerprint data, resulting in poor robustness of the RFID tag model in complex environments; at the same time, these technologies do not combine visual information to improve the positioning accuracy of the RFID tag model.

[0006] Therefore, a precise positioning method and system based on ultra-high frequency RFID electronic tags are proposed. Summary of the invention

[0007] The purpose of the present invention is to provide a precise positioning method and system based on ultra-high frequency RFID electronic tags. First, the fingerprint data of all reference mobile tags are obtained; secondly, the collected fingerprint data is preprocessed, and the first fingerprint data after mean filtering and the second fingerprint data after residual Gaussian filtering are fused using wavelet transform fusion technology, which effectively improves the robustness of the model in complex signal environments; then, a position prediction model based on deep information feature embedding is proposed, which uses a deep information-guided solution to accurately obtain the position information of the mobile tag; finally, a hybrid positioning error is proposed for training the position prediction model, which effectively improves the prediction accuracy of the position prediction model by weighted calculation of fingerprint feature differences and predicted position differences.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A precise positioning method based on ultra-high frequency RFID electronic tags, comprising:

[0010] Obtaining fingerprint data of all reference mobile tags;

[0011] Preprocessing the fingerprint data to obtain preprocessed fingerprint data; wherein the preprocessing includes fusion filtering and normalization;

[0012] The image data obtained by the visual camera is respectively input into the positioning algorithm, the depth prediction model and the target detection model to obtain the posture information, the depth information and the category and position information; wherein the depth prediction model and the target detection model are pre-trained models;

[0013] Integrate the posture information, the depth information and the category information to obtain a first position coordinate;

[0014] Constructing a position prediction model based on embedding of depth information features, inputting the preprocessed fingerprint data into the position prediction model, and obtaining a second position coordinate;

[0015] Optimizing the position prediction model according to the mixed positioning error calculated from the first position coordinates and the second position coordinates to obtain a final position prediction model;

[0016] The fingerprint data of the mobile tag to be located is input into the final position prediction model to obtain the position coordinates of the mobile tag.

[0017] Further, fingerprint data of all reference mobile tags are obtained;

[0018] Further, the fingerprint data is preprocessed to obtain preprocessed fingerprint data; wherein the preprocessing includes fusion filtering and normalization; wherein the preprocessing process includes: obtaining the fingerprint data of the reference positioning point; processing the fingerprint data using fusion filtering to obtain filtered fingerprint data; processing the filtered fingerprint data using normalization to obtain the preprocessed fingerprint data;

[0019] The specific implementation process of the fusion filtering includes:

[0020] Processing the fingerprint data using mean filtering to obtain first fingerprint data;

[0021] Further, the fingerprint data is processed using residual Gaussian filtering to obtain second fingerprint data;

[0022] Further, the first fingerprint data and the second fingerprint data are fused using wavelet transform fusion technology to obtain the filtered fingerprint data;

[0023] The specific implementation process of the fusion includes:

[0024] Using wavelet transform to process the first fingerprint data and the second fingerprint data respectively, so as to obtain high-frequency components and low-frequency components of the two fingerprint data;

[0025] Merging the average values ​​of the high frequency component and the low frequency component of the two fingerprint data;

[0026] The fused fingerprint data is reconstructed by inverse wavelet transform to obtain the filtered fingerprint data.

[0027] Further, the image data obtained by the visual camera is respectively input into the positioning algorithm, the depth prediction model and the target detection model to obtain the posture information, the depth information and the category and position information; wherein the depth prediction model and the target detection model are pre-trained models;

[0028] Further, the posture information, the depth information and the category information are integrated to obtain a first position coordinate;

[0029] Further, constructing a position prediction model based on embedding of depth information features, inputting the preprocessed fingerprint data into the position prediction model, and obtaining a second position coordinate;

[0030] The specific implementation process of the position prediction model based on depth information feature embedding includes:

[0031] The convolutional network applies convolutional layers to the preprocessed data for feature extraction and obtains the initial feature map;

[0032] Furthermore, the encoding end performs multiple downsampling operations on the initial feature map, and before each downsampling operation, multiple feature extraction modules are used to obtain low-dimensional deep-level fingerprint features;

[0033] Furthermore, the decoding end performs multiple upsampling operations on the low-dimensional deep fingerprint features, and the same number of feature extraction modules are used before each operation. At the same time, the convolution features before downsampling are added to the feature graphs of the features after upsampling using skip connections, and then the depth information features are embedded before the skip connections, and finally the second position coordinates of the mobile tag are obtained;

[0034] The process of obtaining the second position coordinates can be expressed by the following formula:

[0035] (P x ,P y )=F c (R i ,D);

[0036] Among them, F c () represents the convolutional neural network based on deep information feature embedding; R i is represented as the preprocessed data; D is represented as the depth information feature; (P x ,P y ) represents the second position coordinate output by the model.

[0037] Further, the position prediction model is optimized according to the mixed positioning error calculated based on the first position coordinates and the second position coordinates to obtain a final position prediction model;

[0038] The specific implementation process of the mixed positioning error includes:

[0039] Obtaining predicted fingerprint features and predicted location coordinates of the location prediction model;

[0040] Further, the real fingerprint features and real position coordinates under the actual label in the fingerprint database are obtained; wherein the fingerprint features include intensity features and phase features;

[0041] Further, the difference between the intensity feature and the phase feature of the two fingerprint features is calculated, and corresponding weights are set to obtain a feature error;

[0042] Furthermore, the Euclidean distance between the two position coordinates is calculated to obtain the position error;

[0043] Further, the feature error and the position error are weightedly summed to obtain the mixed positioning error;

[0044] The calculation formula of the mixed positioning error is as follows:

[0045] E = ω*pe+(1-ω)*fe;

[0046]

[0047] Wherein, M represents the number of fingerprint feature samples; E represents the mixed positioning error; pe represents the position error; fe represents the feature error; Represented as the real position coordinates; Expressed as the predicted position coordinates; qd i It is expressed as the number of intensity feature values ​​of the i-th fingerprint feature sample; It is represented as the j1th true intensity feature value of the i-th fingerprint feature sample; It is represented as the j2th predicted strength feature value of the i-th fingerprint feature sample; xw i It is expressed as the number of phase feature values ​​of the i-th fingerprint feature sample; It is represented as the j2th true phase feature value of the i-th fingerprint feature sample; It is represented as the j2th predicted phase feature value of the i-th fingerprint feature sample; ω and β represent different weights respectively.

[0048] Furthermore, the fingerprint data of the mobile tag to be located is input into the final position prediction model to obtain the position coordinates of the mobile tag.

[0049] A precise positioning system based on ultra-high frequency RFID electronic tags, comprising: a device control module, a data acquisition module, a data processing module, a position prediction module and an output module; wherein the device control module is used to control the start, pause and stop of the device; the data acquisition module is used to obtain image data of a visual camera and fingerprint data of a mobile tag; the data processing module is used to pre-process the acquired fingerprint data to obtain pre-processed fingerprint data; the position prediction module comprises a first position prediction unit and a second position prediction unit; the first position prediction unit is used to perform position prediction on the acquired image data to obtain a first position coordinate; the second position prediction unit is used to perform position prediction on the pre-processed fingerprint data to obtain a second position coordinate; the output module is used to display the final position prediction result of the mobile tag positioning point.

[0050] The specific implementation process of the data processing module includes:

[0051] Receiving the fingerprint data output by the data acquisition module;

[0052] Further, the fingerprint data is processed using fusion filtering to obtain filtered fingerprint data; wherein the specific implementation process of the fusion filtering includes:

[0053] Processing the fingerprint data using mean filtering to obtain first fingerprint data;

[0054] Process the fingerprint data using residual Gaussian filtering to obtain second fingerprint data;

[0055] Using wavelet transform fusion technology to fuse the first fingerprint data and the second fingerprint data to obtain the filtered fingerprint data;

[0056] The specific implementation process of the fusion includes:

[0057] Using wavelet transform to process the first fingerprint data and the second fingerprint data respectively, so as to obtain high-frequency components and low-frequency components of the two fingerprint data;

[0058] Merging the average values ​​of the high frequency component and the low frequency component of the two fingerprint data;

[0059] The fused fingerprint data is reconstructed by inverse wavelet transform to obtain the filtered fingerprint data.

[0060] Furthermore, the filtered fingerprint data is processed using normalization to obtain the pre-processed fingerprint data.

[0061] The second position prediction unit adopts a convolutional neural network based on deep information feature embedding, and the specific implementation process includes:

[0062] The convolutional network applies convolutional layers to the preprocessed data for feature extraction and obtains the initial feature map;

[0063] Furthermore, the encoding end performs multiple downsampling operations on the initial feature map, and before each downsampling operation, multiple feature extraction modules are used to obtain low-dimensional deep-level fingerprint features;

[0064] Furthermore, the decoding end performs multiple upsampling operations on the low-dimensional deep fingerprint features, and the same number of feature extraction modules are used before each operation. At the same time, the convolution features before downsampling are added to the feature graphs of the features after upsampling using skip connections, and then the depth information features are embedded before the skip connections, and finally the second position coordinates of the mobile tag are obtained;

[0065] The process of obtaining the second position coordinates can be expressed by the following formula:

[0066] (P x,P y )=F c (R i ,D);

[0067] Among them, F c () represents the convolutional neural network based on deep information feature embedding; R i is represented as the preprocessed data; D is represented as the depth information feature; (P x ,P y ) represents the second position coordinate output by the model.

[0068] The second position prediction unit is optimized and trained by using a hybrid positioning error, and a specific implementation process of the hybrid positioning error includes:

[0069] Obtaining predicted fingerprint features and predicted location coordinates of the location prediction model;

[0070] Further, the real fingerprint features and real position coordinates under the actual label in the fingerprint database are obtained; wherein the fingerprint features include intensity features and phase features;

[0071] Further, the difference between the intensity feature and the phase feature of the two fingerprint features is calculated, and corresponding weights are set to obtain a feature error;

[0072] Furthermore, the Euclidean distance between the two position coordinates is calculated to obtain the position error;

[0073] Further, the feature error and the position error are weightedly summed to obtain the mixed positioning error;

[0074] The calculation formula of the mixed positioning error is as follows:

[0075] E = ω*pe+(1-ω)*fe;

[0076]

[0077] Wherein, M represents the number of fingerprint feature samples; E represents the mixed positioning error; pe represents the position error; fe represents the feature error; Represented as the real position coordinates; Expressed as the predicted position coordinates; qd i It is expressed as the number of intensity feature values ​​of the i-th fingerprint feature sample; It is represented as the j1th true intensity feature value of the i-th fingerprint feature sample; It is represented as the j2th predicted strength feature value of the i-th fingerprint feature sample; xw i It is expressed as the number of phase feature values ​​of the i-th fingerprint feature sample; It is represented as the j2th true phase feature value of the i-th fingerprint feature sample; It is represented as the j2th predicted phase feature value of the i-th fingerprint feature sample; ω and β represent different weights respectively.

[0078] Compared with the prior art, the present invention has the following beneficial effects:

[0079] 1. The present invention proposes a fusion filtering method. The method first obtains the first fingerprint data and the second fingerprint data respectively through a mean filtering algorithm and a residual Gaussian filtering algorithm, and then fuses the two fingerprint data using a wavelet transform fusion technique to obtain filtered fingerprint data. This can avoid interference to the fingerprint data caused by complex environmental factors such as multipath effects, reflections and refractions, improve the optimization quality of the fingerprint data, and further improve the prediction robustness of the model.

[0080] 2. The present invention proposes a position prediction model based on depth information feature embedding, which embeds the output features of the depth prediction model into the prediction model, and accurately obtains the position information of the mobile tag to be located by using a depth information guided scheme; the model is a convolutional network algorithm based on depth information feature embedding, which embeds depth information features before the decoder downsampling operation, and uses the relative position relationship of different label objects in the depth feature map to help locate the predicted position of the target label, thereby improving the model's ability to predict the position of the mobile label.

[0081] 3. The present invention proposes a hybrid positioning error, which is used to train a position prediction model based on deep information feature embedding; the error first obtains the predicted fingerprint features and predicted position coordinates of the position prediction model and the real fingerprint features and real position coordinates in the fingerprint database, and then is obtained by calculating the weighted sum of the feature error and the position error; this can effectively help the position prediction model learn the mapping relationship between fingerprint data and position coordinates, and further improve the accuracy of the model's position prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 This is a flow chart of a precise positioning method based on ultra-high frequency RFID electronic tags of the present invention;

[0083] Figure 2 A schematic diagram of the framework structure of a precise positioning method based on an ultra-high frequency RFID electronic tag of the present invention;

[0084] Figure 3 A schematic diagram of the network structure of a position prediction model based on embedding of depth information features of the present invention;

[0085] Figure 4It is a structural schematic diagram of a precise positioning system based on ultra-high frequency RFID electronic tags of the present invention. DETAILED DESCRIPTION

[0086] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0087] The current global satellite navigation system can provide people with ready location information in outdoor environments, greatly facilitating people's travel. However, in indoor environments, the non-line-of-sight propagation and multipath effects caused by building occlusion and the complexity of the indoor environment have greatly limited the personnel positioning technology based on satellite navigation.

[0088] At present, RFID technology has been widely used in indoor positioning. When accurately locating RFID tags, it mainly relies on the nonlinear mapping model between signal strength (RSSI) and distance. The basic principle of RSSI positioning technology is that the attenuation of the radio frequency signal is inversely proportional to the square of the distance, and the signal transmission distance is inferred by measuring the power strength of the received signal. However, due to possible interference in the environment, this mapping relationship will fluctuate. Therefore, studying how to quantify the relationship between RSSI and distance has become one of the key issues in solving RFID tag positioning technology.

[0089] The existing technologies perform a single filtering operation on the input fingerprint data, resulting in poor robustness of the RFID tag model in complex environments; at the same time, these technologies do not combine visual information to improve the positioning accuracy of the RFID tag model; for this reason, the present invention proposes a precise positioning method and system based on ultra-high frequency RFID electronic tags; the following will provide a specific description from two embodiments.

[0090] Embodiment 1

[0091] The specific implementation process in the embodiment of the present application is implemented by a precise positioning method based on an ultra-high frequency RFID electronic tag of the present invention, see Figure 1 The process of the method proposed in the present invention is described in the following; the precise positioning method based on ultra-high frequency RFID electronic tags includes:

[0092] S10. Obtain fingerprint data of all reference mobile tags;

[0093] S20. Preprocessing the acquired RFID electronic tag fingerprint data;

[0094] S30. Inputting the image data obtained by the visual camera into the positioning algorithm, the depth prediction model and the target detection model to obtain the first position coordinates;

[0095] S40. Constructing a position prediction model based on embedding of depth information features, inputting the preprocessed fingerprint data into the position prediction model, and obtaining a second position coordinate;

[0096] S50. Optimizing the position prediction model according to the mixed positioning error calculated based on the first position coordinates and the second position coordinates to obtain a final position prediction model;

[0097] S60. Input the fingerprint data of the mobile tag to be located into the final location prediction model to obtain the location coordinates of the mobile tag.

[0098] Furthermore, a specific implementation process of a precise positioning method based on ultra-high frequency RFID electronic tags is as follows:

[0099] Obtaining fingerprint data of all reference mobile tags;

[0100] Preprocessing the fingerprint data to obtain preprocessed fingerprint data; wherein the preprocessing includes fusion filtering and normalization;

[0101] The image data obtained by the visual camera is respectively input into the positioning algorithm, the depth prediction model and the target detection model to obtain the posture information, the depth information and the category and position information; wherein the depth prediction model and the target detection model are pre-trained models;

[0102] Integrate the posture information, the depth information, and the category and position information to obtain a first position coordinate;

[0103] Constructing a position prediction model based on embedding of depth information features, inputting the preprocessed fingerprint data into the position prediction model, and obtaining a second position coordinate;

[0104] Optimizing the position prediction model according to the mixed positioning error calculated from the first position coordinates and the second position coordinates to obtain a final position prediction model;

[0105] The fingerprint data of the mobile tag to be located is input into the final position prediction model to obtain the position coordinates of the mobile tag.

[0106] This example proposes a precise positioning method based on ultra-high frequency RFID electronic tags. The overall framework structure of this method is referenced Figure 2. First, the fingerprint data of all reference mobile tags are obtained; secondly, the collected fingerprint data is preprocessed, and the first fingerprint data after mean filtering and the second fingerprint data after residual Gaussian filtering are fused using wavelet transform fusion technology, which effectively improves the robustness of the model in complex signal environments; then, a position prediction model based on depth information feature embedding is proposed, which can accurately obtain the position information of mobile tags using a depth information-guided scheme; finally, a hybrid positioning error is proposed for training the position prediction model, which effectively improves the prediction accuracy of the position prediction model by weighted calculation of fingerprint feature differences and predicted position differences.

[0107] For specific explanation, the present invention is described in conjunction with the following embodiments, as follows:

[0108] Obtaining fingerprint data of all reference mobile tags;

[0109] In this example, the fingerprint data of the mobile tag includes signal strength data and phase difference data. The preprocessing is optimized for the signal strength data. The location prediction model uses the signal strength data as input data to learn the nonlinear relationship between RSSI and location coordinates, while the hybrid positioning error uses the signal strength data and phase difference data to optimize the location prediction model and improve the positioning accuracy of the location prediction model.

[0110] Further, the fingerprint data is preprocessed to obtain preprocessed fingerprint data; wherein the preprocessing includes fusion filtering and normalization; the specific implementation process of the preprocessing includes: obtaining the fingerprint data of the reference positioning point; processing the fingerprint data using fusion filtering to obtain filtered fingerprint data; processing the filtered fingerprint data using normalization to obtain the preprocessed fingerprint data;

[0111] The specific implementation process of the fusion filtering includes:

[0112] The fingerprint data is processed using mean filtering to obtain first fingerprint data; the calculation formula of the mean filtering is as follows:

[0113]

[0114] Among them, m is the number of acquisitions; Represented as the first fingerprint data; RSSI i It is represented as the signal intensity of the ith acquisition;

[0115] Further, the fingerprint data is processed using a residual Gaussian filter to obtain second fingerprint data; wherein the calculation formula of the residual Gaussian filter is as follows:

[0116]

[0117] Among them, μ is the mean of Gaussian function; σ is the standard deviation of Gaussian function; m is the number of acquisitions; G() is Gaussian filter function; RSSI i It is represented as the signal strength collected for the i-th time; F() is the second fingerprint data output; p is a dynamic factor, which can be dynamically adjusted according to the learning results.

[0118] Further, the first fingerprint data and the second fingerprint data are fused using wavelet transform fusion technology to obtain the filtered fingerprint data;

[0119] The specific implementation process of the fusion includes:

[0120] Using wavelet transform to process the first fingerprint data and the second fingerprint data respectively, so as to obtain high-frequency components and low-frequency components of the two fingerprint data;

[0121] Further, the average values ​​of the high frequency components and the low frequency components of the two fingerprint data are merged;

[0122] Furthermore, the fused fingerprint data is reconstructed by performing inverse wavelet transform to obtain the filtered fingerprint data.

[0123] Among them, the basic principle of using wavelet transform fusion technology in the embodiment of the present application is as follows:

[0124] When using M layers of wavelet transform to decompose and process data, (3M+1) layers of subbands can be obtained, including: low-frequency baseband and layer high-frequency subband D h , D v and D d ;

[0125] When the first fingerprint data and the second fingerprint data are respectively recorded as and When; the proportional coefficient matrix in the filter coefficient matrix is ​​set to H, and the wavelet coefficient matrix is ​​set to G; then the two-dimensional wavelet decomposition algorithm can be described as: Wherein, j represents the number of the decomposition layer; i represents the number of the fingerprint data; h represents the horizontal component; v represents the vertical component; d represents the diagonal component; H Τ It is represented as the conjugate transformed rank matrix of H; G Τ It is represented as the conjugate transformed rank matrix of G;

[0126] Among them, the wavelet transform algorithm is expressed as:

[0127] In addition, in order to further illustrate that the fusion filtering method proposed in the present invention has certain effectiveness in the position prediction model, the algorithm proposed in the present invention is compared with the algorithm with the same function. The comparison results of various evaluation indicators are given in Table 1.

[0128] Table 1. Comparison of location prediction performance of various algorithms

[0129] algorithm Accuracy (%) False alarm rate (%) Detection rate (%) Mean Filter 84.71 14.08 84.62 Gaussian filtering 86.57 13.17 85.63 Residual Gaussian Filter 88.51 12.41 87.04 Fusion Filter 90.08 10.22 89.91

[0130] Table 1 gives the performance comparison data of each algorithm for location prediction, and in order to illustrate the necessity of using a fusion filtering scheme in preprocessing. It can be concluded from the table that when the traditional Gaussian filtering method is used in preprocessing, the accuracy, false alarm rate and detection rate are 86.57%, 13.17% and 85.63% respectively, and the model performance is significantly lower than the model using the hybrid filtering method; therefore, it can be seen that the hybrid filtering method can significantly improve the prediction performance of the location prediction model.

[0131] The preprocessing process in this case includes fusion filtering and normalization; the process first obtains the fingerprint data of the reference positioning point; then uses fusion filtering to process the fingerprint data to obtain filtered fingerprint data; then uses normalization to process the filtered fingerprint data to obtain the preprocessed fingerprint data; by preprocessing the fingerprint data, the position prediction capability of the position prediction model can be improved.

[0132] In this example, the fusion filtering method first obtains the first fingerprint data and the second fingerprint data respectively through the mean filtering algorithm and the residual Gaussian filtering algorithm, and then fuses the two fingerprint data using the wavelet transform fusion technology to obtain the filtered fingerprint data. This can avoid the interference of complex environmental factors such as multipath effects, reflection and refraction on the fingerprint data, improve the optimization quality of the fingerprint data, and further improve the prediction robustness of the model.

[0133] Further, the image data obtained by the visual camera is respectively input into the positioning algorithm, the depth prediction model and the target detection model to obtain the posture information, the depth information and the category and position information; wherein the depth prediction model and the target detection model are pre-trained models;

[0134] Among them, the positioning algorithm adopts an instant positioning and map construction algorithm, which extracts ORB features in the image data and matches the number of feature points to calculate the pose information (R, t); the depth prediction model is a convolutional neural network, which converts image features into depth information features D; the target detection model uses the YoloV5 model, which is used to identify the location information Pr of the label.

[0135] Further, integrating the posture information, the depth information, and the category and position information to obtain a first position coordinate;

[0136] Among them, the specific implementation process of the first position coordinate is to first obtain the real three-dimensional coordinate Pw of the detection object through the formula D*Pr=K*(R*Pw+t), K is the camera internal parameter; then project the three-dimensional coordinate Pw into the two-dimensional coordinate system to obtain the first position coordinate.

[0137] Further, constructing a position prediction model based on embedding of depth information features, inputting the preprocessed fingerprint data into the position prediction model, and obtaining a second position coordinate;

[0138] The position prediction model is a convolutional neural network based on deep information feature embedding. The network structure of the model is referenced Figure 3 The specific implementation process includes:

[0139] The convolutional network applies convolutional layers to the preprocessed data for feature extraction and obtains the initial feature map;

[0140] Furthermore, the encoding end performs multiple downsampling operations on the initial feature map, and before each downsampling operation, multiple feature extraction modules are used to obtain low-dimensional deep-level fingerprint features;

[0141] Furthermore, the decoding end performs multiple upsampling operations on the low-dimensional deep fingerprint features, and the same number of feature extraction modules are used before each operation. At the same time, the convolution features before downsampling are added to the feature graphs of the features after upsampling using skip connections, and then the depth information features are embedded before the skip connections, and finally the second position coordinates of the mobile tag are obtained;

[0142] The process of obtaining the second position coordinates can be expressed by the following formula:

[0143] (P x ,P y )=F c (R i ,D);

[0144] Among them, F c () represents the convolutional neural network based on deep information feature embedding; R i is represented as the preprocessed data; D is represented as the depth information feature; (P x ,P y ) represents the second position coordinate output by the model.

[0145] In addition, in order to further illustrate that the depth information feature embedding proposed in the present invention plays a certain effectiveness in position prediction, the model proposed in the present invention is compared with the model with the same function. The comparison results of various evaluation indicators are given in Table 2.

[0146] Table 2. Comparison of location prediction performance of each model

[0147]

[0148] Table 2 gives the performance comparison data of each model for location prediction, and in order to illustrate the necessity of embedding deep information features in the location prediction model. It can be concluded from the table that when the location prediction model does not embed deep information features, the accuracy, false alarm rate and detection rate are 88.15%, 11.09% and 88.06% respectively, and the model performance is significantly lower than the location prediction model with deep information features embedded; therefore, it can be seen that deep information feature embedding can significantly improve the performance of the location prediction model.

[0149] In this example, a position prediction model based on deep information feature embedding is used to map input fingerprint data into precise position coordinates; the model embeds the output features of the deep prediction model into the prediction model, and accurately obtains the position information of the mobile tag to be located by using a depth information-guided scheme; the model is a convolutional network algorithm based on deep information feature embedding, which embeds deep information features before the decoder downsampling operation, and uses the relative position relationship of different label objects in the deep feature map to help locate the predicted position of the target label, thereby improving the model's ability to predict the position of the mobile tag.

[0150] Further, the position prediction model is optimized according to the mixed positioning error calculated based on the first position coordinates and the second position coordinates to obtain a final position prediction model;

[0151] The specific implementation process of the mixed positioning error includes:

[0152] Obtaining predicted fingerprint features and predicted location coordinates of a location prediction model;

[0153] Further, the real fingerprint features and real position coordinates under the actual label in the fingerprint database are obtained; wherein the fingerprint features include intensity features and phase features;

[0154] Further, the difference between the intensity feature and the phase feature of the two fingerprint features is calculated, and corresponding weights are set to obtain a feature error;

[0155] Furthermore, the Euclidean distance between the two position coordinates is calculated to obtain the position error;

[0156] Further, the feature error and the position error are weightedly summed to obtain the mixed positioning error;

[0157] The calculation formula of the mixed positioning error is as follows:

[0158] E = ω*pe+(1-ω)*fe;

[0159]

[0160] Wherein, M represents the number of fingerprint feature samples; E represents the mixed positioning error; pe represents the position error; fe represents the feature error; Represented as the real position coordinates; Expressed as the predicted position coordinates; qd i It is expressed as the number of intensity feature values ​​of the i-th fingerprint feature sample; It is represented as the j1th true intensity feature value of the i-th fingerprint feature sample; It is represented as the j2th predicted strength feature value of the i-th fingerprint feature sample; xw i It is expressed as the number of phase feature values ​​of the i-th fingerprint feature sample; It is represented as the j2th true phase feature value of the i-th fingerprint feature sample; It is represented as the j2th predicted phase feature value of the ith fingerprint feature sample; ω and β represent different weights, which are 0.5 and 0.6 respectively.

[0161] In this case, a hybrid positioning error is proposed, which is used to train a location prediction model based on deep information feature embedding. The error first obtains the predicted fingerprint features and predicted location coordinates of the location prediction model and the real fingerprint features and real location coordinates in the fingerprint database, and then calculates the weighted sum of the feature error and the location error. This can effectively help the location prediction model learn the mapping relationship between fingerprint data and location coordinates, and further improve the accuracy of the model's location prediction.

[0162] Furthermore, the fingerprint data of the mobile tag to be located is input into the final position prediction model to obtain the position coordinates of the mobile tag.

[0163] Embodiment 2

[0164] As an embodiment of the present invention, refer to Figure 4 ,A precise positioning system based on UHF RFID electronic tag includes: a device control module, a data acquisition module, a data processing module, a location prediction module and an output module;

[0165] Wherein, the device control module is used to control the start, pause and stop of the device;

[0166] The data acquisition module is used to obtain image data from the visual camera and fingerprint data from the mobile tag;

[0167] The data processing module is used to preprocess the acquired fingerprint data to obtain preprocessed fingerprint data; wherein the preprocessing process includes: obtaining the fingerprint data of the reference positioning point; processing the fingerprint data using fusion filtering to obtain filtered fingerprint data; processing the filtered fingerprint data using normalization to obtain the preprocessed fingerprint data;

[0168] The specific implementation process of the fusion filtering includes:

[0169] Processing the fingerprint data using mean filtering to obtain first fingerprint data;

[0170] Further, the fingerprint data is processed using residual Gaussian filtering to obtain second fingerprint data;

[0171] Furthermore, the first fingerprint data and the second fingerprint data are fused using a wavelet transform fusion technology to obtain the filtered fingerprint data.

[0172] The position prediction module includes a first position prediction unit and a second position prediction unit; wherein the first position prediction unit is used to perform position prediction on the acquired image data to obtain a first position coordinate; and the second position prediction unit is used to perform position prediction on the pre-processed fingerprint data to obtain a second position coordinate;

[0173] The second position prediction unit adopts a convolutional neural network based on deep information feature embedding, and the specific implementation process includes:

[0174] The convolutional network applies convolutional layers to the preprocessed data for feature extraction and obtains the initial feature map;

[0175] Furthermore, the encoding end performs multiple downsampling operations on the initial feature map, and before each downsampling operation, multiple feature extraction modules are used to obtain low-dimensional deep-level fingerprint features;

[0176] Furthermore, the decoding end performs multiple upsampling operations on the low-dimensional deep fingerprint features, and the same number of feature extraction modules are used before each operation. At the same time, the convolution features before downsampling are added to the feature graphs of the features after upsampling using skip connections, and then the depth information features are embedded before the skip connections, and finally the second position coordinates of the mobile tag are obtained;

[0177] The process of obtaining the second position coordinates can be expressed by the following formula:

[0178] (Px ,P y )=F c (R i ,D);

[0179] Among them, F c () represents the convolutional neural network based on deep information feature embedding; R i is represented as the preprocessed data; D is represented as the depth information feature; (P x ,P y ) represents the second position coordinate output by the model.

[0180] The second position prediction unit is optimized and trained by using a hybrid positioning error, and a specific implementation process of the hybrid positioning error includes:

[0181] Obtaining predicted fingerprint features and predicted location coordinates of the location prediction model;

[0182] Further, the real fingerprint features and real position coordinates under the actual label in the fingerprint database are obtained; wherein the fingerprint features include intensity features and phase features;

[0183] Further, the difference between the intensity feature and the phase feature of the two fingerprint features is calculated, and corresponding weights are set to obtain a feature error;

[0184] Furthermore, the Euclidean distance between the two position coordinates is calculated to obtain the position error;

[0185] Further, the feature error and the position error are weightedly summed to obtain the mixed positioning error;

[0186] The calculation formula of the mixed positioning error is as follows:

[0187] E = ω*pe+(1-ω)*fe;

[0188]

[0189] Wherein, M represents the number of fingerprint feature samples; E represents the mixed positioning error; pe represents the position error; fe represents the feature error; Represented as the real position coordinates; Expressed as the predicted position coordinates; qd i It is expressed as the number of intensity feature values ​​of the i-th fingerprint feature sample; It is represented as the j1th true intensity feature value of the i-th fingerprint feature sample; It is represented as the j2th predicted strength feature value of the i-th fingerprint feature sample; xw iIt is expressed as the number of phase feature values ​​of the i-th fingerprint feature sample; It is represented as the j2th true phase feature value of the i-th fingerprint feature sample; It is represented as the j2th predicted phase feature value of the i-th fingerprint feature sample; ω and β represent different weights respectively.

[0190] The output module is used to display the final position prediction result of the mobile tag positioning point.

[0191] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A precise positioning method based on ultra-high frequency RFID electronic tags, It is characterized in that include: Obtaining fingerprint data of all reference mobile tags; Preprocessing the fingerprint data to obtain preprocessed fingerprint data; wherein the preprocessing includes fusion filtering and normalization; The image data obtained by the visual camera is respectively input into the positioning algorithm, the depth prediction model and the target detection model to obtain the posture information, the depth information and the category and position information; wherein the depth prediction model and the target detection model are pre-trained models; Integrate the posture information, the depth information and the category information to obtain a first position coordinate; Constructing a position prediction model based on embedding of depth information features, inputting the preprocessed fingerprint data into the position prediction model, and obtaining a second position coordinate; Optimizing the position prediction model according to the mixed positioning error calculated from the first position coordinates and the second position coordinates to obtain a final position prediction model; The fingerprint data of the mobile tag to be located is input into the final position prediction model to obtain the position coordinates of the mobile tag.

2. According to claim 1, a precise positioning method based on ultra-high frequency RFID electronic tags, It is characterized in that The pretreatment process includes: Obtaining fingerprint data of the reference positioning point; Process the fingerprint data using fusion filtering to obtain filtered fingerprint data; The filtered fingerprint data is processed using normalization to obtain the pre-processed fingerprint data.

3. According to claim 2, a precise positioning method based on ultra-high frequency RFID electronic tags, It is characterized in that The specific implementation process of the fusion filtering includes: Processing the fingerprint data using mean filtering to obtain first fingerprint data; Process the fingerprint data using residual Gaussian filtering to obtain second fingerprint data; Using wavelet transform fusion technology to fuse the first fingerprint data and the second fingerprint data to obtain the filtered fingerprint data; The specific implementation process of the fusion includes: Using wavelet transform to process the first fingerprint data and the second fingerprint data respectively, so as to obtain high-frequency components and low-frequency components of the two fingerprint data; Merging the average values ​​of the high frequency component and the low frequency component of the two fingerprint data; The fused fingerprint data is reconstructed by inverse wavelet transform to obtain the filtered fingerprint data.

4. According to claim 1, a precise positioning method based on ultra-high frequency RFID electronic tags, It is characterized in that The position prediction model is a convolutional neural network based on deep information feature embedding, and the specific implementation process includes: The convolutional network applies convolutional layers to the preprocessed data for feature extraction and obtains the initial feature map; The encoding end performs multiple downsampling operations on the initial feature map, and before each downsampling operation, multiple feature extraction modules are used to obtain low-dimensional deep-level fingerprint features; The decoding end performs multiple upsampling operations on the low-dimensional deep fingerprint features, and the same number of feature extraction modules are used before each operation. At the same time, the convolution features before downsampling are added to the feature graphs of the features after upsampling using skip connections, and then the depth information features are embedded before the skip connections, and finally the second position coordinates of the mobile tag are obtained; The process of obtaining the second position coordinates can be expressed by the following formula: (P x ,P y )=F c (R i ,D); Among them, F c () represents the convolutional neural network based on deep information feature embedding; R i is represented as the preprocessed data; D is represented as the depth information feature; (P x ,P y ) represents the second position coordinate output by the model.

5. According to claim 1, a precise positioning method based on ultra-high frequency RFID electronic tags, It is characterized in that The specific implementation process of the hybrid positioning error includes: Obtaining predicted fingerprint features and predicted location coordinates of the location prediction model; Obtaining the real fingerprint features and real position coordinates under the actual tag in the fingerprint database; wherein the fingerprint features include intensity features and phase features; Calculating the difference between the intensity feature and the phase feature of the two fingerprint features, and setting corresponding weights to obtain a feature error; Calculate the Euclidean distance between the two position coordinates to obtain the position error; Performing weighted summation on the feature error and the position error to obtain the mixed positioning error; The calculation formula of the mixed positioning error is as follows: E = ω*pe+(1-ω)*fe; Wherein, M represents the number of fingerprint feature samples; E represents the mixed positioning error; pe represents the position error; fe represents the feature error; Represented as the real position coordinates; Expressed as the predicted position coordinates; qd i It is expressed as the number of intensity feature values ​​of the i-th fingerprint feature sample; Represented as the jth sample of the i-th fingerprint feature 1 True intensity eigenvalues; Represented as the jth sample of the i-th fingerprint feature 2 The predicted strength feature value; xw i It is expressed as the number of phase feature values ​​of the i-th fingerprint feature sample; Represented as the jth sample of the i-th fingerprint feature 2 True phase eigenvalues; Represented as the jth sample of the i-th fingerprint feature 2 predicted phase eigenvalues; ω and β represent different weights respectively.

6. A precise positioning system based on ultra-high frequency RFID electronic tags, It is characterized in that include: Equipment control module, data acquisition module, data processing module, position prediction module and output module; wherein, the equipment control module is used to control the start, pause and stop of the equipment; the data acquisition module is used to obtain the image data of the visual camera and the fingerprint data of the mobile tag; the data processing module is used to pre-process the acquired fingerprint data to obtain pre-processed fingerprint data; the position prediction module includes a first position prediction unit and a second position prediction unit; the first position prediction unit is used to perform position prediction on the acquired image data to obtain a first position coordinate; the second position prediction unit is used to perform position prediction on the pre-processed fingerprint data to obtain a second position coordinate; wherein, the second position prediction unit is optimized and trained using a mixed positioning error; the output module is used to display the final position prediction result of the mobile tag positioning point.

7. The precise positioning system based on ultra-high frequency RFID electronic tags according to claim 6, It is characterized in that The specific implementation process of the data processing module includes: Receiving the fingerprint data output by the data acquisition module; Process the fingerprint data using fusion filtering to obtain filtered fingerprint data; The filtered fingerprint data is processed using normalization to obtain the pre-processed fingerprint data.

8. According to claim 7, a precise positioning system based on ultra-high frequency RFID electronic tags, It is characterized in that The specific implementation process of the fusion filtering includes: Processing the fingerprint data using mean filtering to obtain first fingerprint data; Process the fingerprint data using residual Gaussian filtering to obtain second fingerprint data; Using wavelet transform fusion technology to fuse the first fingerprint data and the second fingerprint data to obtain the filtered fingerprint data; The specific implementation process of the fusion includes: Using wavelet transform to process the first fingerprint data and the second fingerprint data respectively, so as to obtain high-frequency components and low-frequency components of the two fingerprint data; Merging the average values ​​of the high frequency component and the low frequency component of the two fingerprint data; The fused fingerprint data is reconstructed by inverse wavelet transform to obtain the filtered fingerprint data.

9. The precise positioning system based on ultra-high frequency RFID electronic tags according to claim 6, It is characterized in that The second position prediction unit adopts a convolutional neural network based on deep information feature embedding, and the specific implementation process includes: The convolutional network applies convolutional layers to the preprocessed data for feature extraction and obtains the initial feature map; The encoding end performs multiple downsampling operations on the initial feature map, and before each downsampling operation, multiple feature extraction modules are used to obtain low-dimensional deep-level fingerprint features; The decoding end performs multiple upsampling operations on the low-dimensional deep fingerprint features, and the same number of feature extraction modules are used before each operation. At the same time, the convolution features before downsampling are added to the feature graphs of the features after upsampling using skip connections, and then the depth information features are embedded before the skip connections, and finally the second position coordinates of the mobile tag are obtained; The process of obtaining the second position coordinates can be expressed by the following formula: (P x ,P y )=F c (R i ,D); Among them, F c () represents the convolutional neural network based on deep information feature embedding; R i is represented as the preprocessed data; D is represented as the depth information feature; (P x ,P y ) represents the second position coordinate output by the model.

10. The precise positioning system based on ultra-high frequency RFID electronic tags according to claim 6, It is characterized in that The specific implementation process of the hybrid positioning error includes: Obtaining predicted fingerprint features and predicted location coordinates of a location prediction model; Obtaining the real fingerprint features and real position coordinates under the actual tag in the fingerprint database; wherein the fingerprint features include intensity features and phase features; Calculating the difference between the intensity feature and the phase feature of the two fingerprint features, and setting corresponding weights to obtain a feature error; Calculate the Euclidean distance between the two position coordinates to obtain the position error; Performing weighted summation on the feature error and the position error to obtain the mixed positioning error; The calculation formula of the mixed positioning error is as follows: E = ω*pe+(1-ω)*fe; Wherein, M represents the number of fingerprint feature samples; E represents the mixed positioning error; pe represents the position error; fe represents the feature error; Represented as the real position coordinates; Expressed as the predicted position coordinates; qd i It is expressed as the number of intensity feature values ​​of the i-th fingerprint feature sample; Represented as the jth sample of the i-th fingerprint feature 1 True intensity eigenvalues; Represented as the jth sample of the i-th fingerprint feature 2 The predicted strength feature value; xw i It is expressed as the number of phase feature values ​​of the i-th fingerprint feature sample; Represented as the jth sample of the i-th fingerprint feature 2 True phase eigenvalues; Represented as the jth sample of the i-th fingerprint feature 2 predicted phase eigenvalues; ω and β represent different weights respectively.

Citation Information

Patent Citations

  • A Deep Learning-Based Indoor 3D Positioning Method for RFID

    CN111537952B

  • A Deep Learning-Based RFID Tag Positioning System

    CN114710831B

  • A multi-sensor fusion positioning method based on RFID

    CN115103299B