A three-dimensional target perception method based on mobile RFID reader and double tags

Abstract

This invention discloses a high-precision 3D positioning algorithm based on a single mobile RFID reader and dual tags on a target. First, the movement of the RFID reader collects the RSS and phase information reflected by a set of dual tags affixed to the target. Second, the peak RSS value of the tags is obtained from the filtered RSS information, and this peak value is multiplied by the robot's velocity to obtain a rough position of the tags relative to the shelf length. Subsequently, based on the preprocessed phase information, the rough position of the target along the shelf depth and height is estimated. Finally, the obtained target position information on the shelf is used as the initial value for model estimation, and the proposed MRRDT algorithm is used to further calibrate the target position information to obtain precise 3D position information. The proposed algorithm achieves centimeter-level accuracy, far exceeding existing methods.

Description

Technical Field

[0001] This invention relates to a three-dimensional target perception method, and more particularly to a three-dimensional target perception method based on a mobile RFID reader and dual tags. Background Technology

[0002] RFID (Radio Frequency Identification) signal-based positioning systems primarily estimate the target's location by utilizing the signal information reflected from the tag attached to the target, received by a reader. Based on the different signal types used, RFID methods are mainly divided into two types: those based on signal strength and those based on signal phase measurement. Methods based on Received Signal Strength (RSS) measurement are further divided into two types: RSS fingerprinting and RSS ranging based on path loss models.

[0003] For methods that convert RSS information into distance information based on electromagnetic wave transmission loss models for ranging and positioning, Ni et al. pioneered the LANDMARC system in their paper (LMNi, Yunhao Liu, Yiu Cho Lau and APPatil, "LANDMARC: indoor location sensing using active RFID," Proceedings of the First IEEE International Conference on Pervasive Computing and Communications, 2003. (PerCom 2003)., 2003, pp. 407-415, doi:10.1109 / PERCOM.2003.1192765). This system deploys a series of RFID tags in the area to be located and performs positioning based on a series of RSS measurements received by the reader. However, RSS information is susceptible to environmental interference, leading to instability, and a high density of reference tags is required to achieve higher positioning accuracy.

[0004] The literature (PV Nikitin, R. Martinez, S. Ramamurthy, H. Leland, G. Spiess and K. V.S. Rao, "Phase based spatial identification of UHF RFID tags," 2010 IEEE International Conference on RFID (IEEE RFID 2010), 2010, pp. 102-109, doi:10.1109 / RFID.2010.5467253) proposes the STPP algorithm, which uses the phase change caused by the distance change between the reader antenna and the tag to determine the relative position of these tags. This method can only obtain the relative position between targets, and its accuracy is not high enough; to obtain higher accuracy, additional equipment deployment or calibration is required.

[0005] In addition, positioning based on the phase difference converted into range difference using RIFD has gradually become a research hotspot. The literature (T. Liu, L. Yang, Q. Lin, Y. Guo and Y. Liu, "Anchor-free backscatterpositioning for RFID tags with high accuracy," IEEE INFOCOM 2014-IEEE Conference on Computer Communications, 2014, pp. 379-387, doi:10.1109 / INFOCOM.2014.6847960) measured the phase difference of the tag relative to the two antennas by deploying dual antennas, and then combined this with the range difference to obtain an average positioning accuracy of 12.8 cm. The paper (ARChatzistefanou and AGDimitriou, "TagLocalization by Handheld UHF RFID Reader and Optical Markers," 2022 IEEE 12th International Conference on RFID Technology and Applications (RFID-TA), 2022, pp.9-12, doi:10.1109 / RFID-TA54958.2022.9924090) proposes to use Kalman filtering to process the image sequence acquired by the camera to obtain the reader's trajectory, and then combine it with the phase difference measurement value collected from the tag at an unknown location to realize the three-dimensional estimation of the tag. The paper (X. Liu et al., "Accurate Localization of Tagged Objects Using Mobile RFID-Augmented Robots," in IEEE Transactions on Mobile Computing, vol. 20, no. 4, pp. 1273-1284, April 1, 2021, doi:10.1109 / TMC.2019.2962129) implements a mobile RFID robot localization (MRL) system. A reader equipped with two vertically deployed antennas continuously acquires the phase difference and time information of tags on shelves during linear movement in a warehouse aisle. By utilizing the geometric relationship between the antenna trajectory and the target tag, high accuracy in two-dimensional and three-dimensional localization is achieved. However, in the three-dimensional MRL system, the localization accuracy in the y- and z-dimensional dimensions is relatively low.

[0006] Combining phase and signal strength is also a promising direction in RFID positioning. The literature (S. Zhang, Y. Fu, D. Jiang and X. Liu, "RFID Localization Based on Multiple Feature Fusion," 2018 15th Annual IEEE International Conference on Sensing, Communication, and Networking (SECON), 2018, pp. 1-2, doi:10.1109 / SAHCN.2018.8397148) first uses RSS to quickly narrow down the possible area of ​​the target tag, and then uses phase information to refine the location estimation. However, this requires deploying a large number of readers and antennas to achieve a small grid and centimeter-level positioning accuracy. Summary of the Invention

[0007] Purpose of the invention: The purpose of this invention is to provide a three-dimensional target perception method based on a mobile RFID reader and dual tags that can reduce computational complexity and achieve higher positioning accuracy.

[0008] Technical solution: The three-dimensional high-precision target perception method of the present invention includes the following steps:

[0009] S1, The RFID reader is mounted on the robot, and the robot's movement collects the RSS and phase information reflected by a set of tags pasted on the target being sensed; and the RSS and phase information are filtered.

[0010] S2. Obtain the peak time of the tag's RSS based on the filtered RSS information, multiply this peak time by the robot's speed to obtain the tag's rough position information relative to the shelf length direction x dimension; then, estimate the rough position information of the perceived target along the shelf depth direction y dimension and height direction z dimension based on the preprocessed phase information.

[0011] S3 uses the obtained length, height, and depth information of the target to be measured as initial values, and further calibrates the position information of the perceived target based on the MRRDT algorithm to obtain accurate position information.

[0012] Furthermore, in step S1, the total time period of the robot's movement is sampled N times, and the label T is set to... a In t n The RSS value obtained from the i-th antenna at time t is rss a,i (t n ), tag T b In t n The RSS value obtained from the i-th antenna at time t is rssb,i (t n ); where i = 1, 2, corresponding to the upper and lower antennas of the reader, respectively; n ∈ [1, N], then all RSS values ​​collected during the total motion time of the robot are obtained, forming a matrix as follows:

[0013] RSS = [rss] a,1 rss b,1 rss a,2 rss b,2 ] T

[0014] in,

[0015] rss a,i =[rss a,i (t1),...,rss a,i (t n ),...,rss a,i (t N )]

[0016] rss b,i =[rss b,i (t1),...,rss b,i (t n ),...,rss b,i (t N )]

[0017]

[0018]

[0019] Where S represents the number of samples, t n,s Indicates at time t n The RSS obtained from the s-th sampling point;

[0020] Define the signal strength difference between tag a and tag b as disRSS. a,b Then we have:

[0021]

[0022] Among them, rss a (t n ) for in t n The time reader receives the signal strength from tag a, rss b (t n ) for in t n The time reader receives the signal strength from tag b.

[0023] Furthermore, in step S1, for time t nThe acquired signal strength value rss(t) n Gaussian filtering is performed, and then the average of multiple sets of values ​​obtained at the same time is taken for smoothing. The resulting RSS matrix is ​​RSS′, which is represented as follows:

[0024] RSS′=[rss′ a,1 rss′ b,1 rss′ a,2 rss′ b,2 ] T

[0025] in,

[0026] rss′ a,i =[rss' a,i (t1),...,rss′ a,i (t n ),...,rss′ a,i (t N )]

[0027] rss′ b,i =[rss' b,i (t1),...,rss′ b,i (t n ),...,rss′ b,i (t N )]

[0028] Where i = 1 and 2, corresponding to the upper and lower antennas of the reader, respectively.

[0029] Furthermore, in step S2, the initial position estimation of the target in the x-dimensional dimension is implemented as follows:

[0030] Let the times corresponding to the peak values ​​of RSS′ be t. a,1 t a,2 t b,1 t b,2 , let t ⊥ Let the moment when the target is closest to the antenna be represented by:

[0031]

[0032] Obtain the initial position estimate of the target in the x-dimensional dimension. Represented as:

[0033]

[0034] Where v represents the robot's moving speed.

[0035] Furthermore, in step S2, a dual-label phase model is used to estimate the position of the target in the y and z dimensions. The specific implementation steps are as follows:

[0036] S21, the target to be tested is labeled with two tags, T, which are attached to the top and bottom respectively. a and T b The distance between the two labels is d a,b d a,i (t n ),d b,i (t n ) represent time t respectively n The i-th antenna of the reader and the tag T a and T b The distance between them is:

[0037]

[0038] in, They represent t respectively n Moment tag T a T b The phase calibration value between the i-th antenna of the reader and the reader, where λ represents the wavelength of the electromagnetic wave emitted by the reader;

[0039] S22, Antenna A i At every moment of the entire process with label T a T b The corresponding distance vector d a,i d b,i They are represented as follows:

[0040] d a,i =[d a,i (t1),...,d a,i (t n ),...,d a,i (t N )]

[0041] d b,i =[d b,i (t1),...,d b,i (t n ),...,d b,i (t N )]

[0042] In t n At time, A i The location is A i (t n ), passing through point A i (t n ) as label T a T bThe perpendicular line to the line intersects at point G. i So, label T a and A i (t n ) and G i The three points can form a right triangle, and the label T b and A i (t n ) and G i The three points can also form a right triangle; therefore, at time t n have:

[0043]

[0044] in, Indicates t n Moment tag T b and projection point G i The distance between them;

[0045] Then we have:

[0046]

[0047] The positional relationship of the target under test is used to estimate the target's z-dimensional position at this moment. T,i (t n The following equation must be satisfied:

[0048]

[0049] Then estimate the z-dimensional value of the target at all times. T,i Taking the average, we have:

[0050]

[0051] At the same time, in label T b and A i (t n ) and G i The right triangle formed by the three points contains:

[0052]

[0053] in, For t n Point G at time i With antenna A i (t n The distance between them;

[0054] However, the true position in the y-dimensional dimension is only when the antenna is perpendicular to the label t. ⊥ We obtain the value at each step, and let the actual value of y be... Solving t in conjunction with the entire process nThe y-value corresponding to time t is:

[0055]

[0056] In the formula t ⊥ This indicates the moment when the target is closest to the antenna, and v represents the robot's moving speed;

[0057] Then for y(t) n ) Calculate the average value of the entire process. Among them, t n ∈[t1,t N ], then we have:

[0058]

[0059] The two sets of data obtained from the upper antenna A1 and the lower antenna A2 Taking the average, we have:

[0060]

[0061] Finally, the initial coarse estimate of the target's position is obtained.

[0062] Furthermore, in step S3, a calibration algorithm based on Taylor series expansion is used to adjust the initial position estimate. Further calibration, specifically implemented through the following steps:

[0063] The entire time process t N It is divided into three equal parts, namely [t1,...,t] w ]、[t w+1 ,...,t 2w ]、[t 2w+1 ,...,t N ], where N = 3w; let t n t n+w and t n+2w The positions of the upper antenna A1 at the three time points are R1, E1, and J1, respectively, and the positions of the upper antenna A2 at the same time points are R2, E2, and J2, respectively; tag T a Points R1, E1, and J1 form a plane. Let label T. a The projection point of the line containing points R1, E1, and J1 is M1. Therefore, points R1, E1, and J1 are respectively aligned with label T. a Points M1 and M2 form three right triangles. and Therefore:

[0064]

[0065] Then from label Ta There is a correlation between the phase and distance of backscattering at different times, which yields:

[0066]

[0067] In X a,1 A1(t1)Y a,1 In a plane, label T a The position is represented as (x a,1 ,y a,1 ), combined with the label T a The positional relationship between the three-dimensional calibration coordinate system and the two-dimensional planar coordinate system is obtained as follows:

[0068]

[0069] Due to velocity v and time point t n t n+w t n+2w Given that, we have:

[0070]

[0071] Combining the above four equations, we get:

[0072]

[0073] A Taylor series expansion is performed, and coarse estimates are made of the x and y dimensions of the target object. As an initial value, we get:

[0074]

[0075] in,

[0076]

[0077] Get label T a X in a two-dimensional plane a,1 Axis, Y a,1 The projections of the axes are x a,1 y a,1 , respectively corresponding in, The label T for the target to be tested a The true position of the x-axis;

[0078] Similarly, label T a The plane formed by points R2, E2, and J2 is obtained. in, The label T for the target to be tested a The true position of the x-axis;

[0079] The obtained xa,1 x a,2 For the estimated label T a Taking the average value at the x-axis position, we have:

[0080]

[0081] Where, x a For label T a Estimate the true x-axis position;

[0082] Let the lines containing M1 and M2 intersect at point o on the x-axis. M Let the line containing M1 and M2 be z. M Axis, through o M The line parallel to the y-axis is y M Axis; in the yoz plane coordinate system of the calibration algorithm, label T a At this time in z M The axis projection points are denoted as p, M2 and label T. a Forming a right triangle and p, M1 and label T a Forming a right triangle, we get:

[0083]

[0084] in, Let p be the distance between points M1; finally, we get:

[0085]

[0086] Then the three-dimensional coordinates (x a ,y a ,z a ) for t n t n+w t n+2w Labels T obtained through calibration algorithms at three time points a 3D coordinates;

[0087] Take the reciprocal of the variance of the estimated values ​​in the x-dimension. The x-dimensional coordinates are calibrated using weights, resulting in the updated x-dimensional coordinates of the target object:

[0088]

[0089] Where, x a,fin For the updated tag T a x-dimensional coordinates, x b,fin For the updated tag T b x-dimensional coordinates;

[0090] Finally, the estimated final three-dimensional coordinate position of the target is obtained as follows:

[0091] Compared with the prior art, the significant advantages of this invention are as follows:

[0092] 1. A closed-form solution for sensing the height of a target relative to a shelf based on a dual-label phase model was derived: First, the phase obtained from the dual labels was converted into a distance model; then, using the geometric relationship between the dual labels and the reader, the functional relationship satisfied by the height of the target relative to the shelf was obtained; the average error of the height estimation was low, reaching centimeter-level accuracy.

[0093] 2. A solution combining three-dimensional coarse position estimation and position calibration is proposed. Initial estimation is performed by combining RSS peak information and dual-label phase model respectively, followed by x-dimensional calibration to correct the initial x-dimensional estimate. Compared with existing positioning algorithms, the present invention has low computational complexity and can achieve higher positioning accuracy, with a positioning error of about 4cm. Attached Figure Description

[0094] Figure 1 This is a schematic diagram of the system architecture of the present invention;

[0095] Figure 2 This is a flowchart of the present invention;

[0096] Figure 3 A schematic diagram illustrating the changes in RSS measurements when the reader and tag positions are fixed.

[0097] Figure 4 A diagram illustrating the peak RSS feed times for different tags;

[0098] Figure 5 This is a schematic diagram of a dual-label phase model;

[0099] Figure 6 This is a schematic diagram of a three-dimensional calibration algorithm;

[0100] Figure 7 A two-dimensional planar diagram of the calibration algorithm;

[0101] Figure 8 This is a schematic diagram of the yoz profile for the calibration algorithm;

[0102] Figure 9(a) shows the X-axis positioning error of the algorithm proposed in this invention and the MRL algorithm in existing literature as a function of the target depth in the shelf.

[0103] Figure 9(b) shows the Y-axis positioning error of the algorithm proposed in this invention and the MRL algorithm in existing literature as a function of the target depth in the shelf.

[0104] Figure 9(c) shows the Z-axis positioning error of the algorithm proposed in this invention and the MRL algorithm as a function of the target depth in the shelf.

[0105] Figure 10(a) shows the X-axis positioning error of the algorithm proposed in this invention and the MRL algorithm as a function of the target's height in the shelf.

[0106] Figure 10(b) shows the Y-axis positioning error of the algorithm proposed in this invention and the MRL algorithm as a function of the target's height in the shelf.

[0107] Figure 10(c) shows the Z-axis positioning error of the algorithm proposed in this invention and the MRL algorithm as a function of the target's height in the shelf.

[0108] Figure 11(a) shows the X-axis positioning error of the algorithm proposed in this invention and the MRL algorithm as a function of the target length in the shelf.

[0109] Figure 11(b) shows the Y-axis positioning error of the algorithm proposed in this invention and the MRL algorithm as a function of the target length in the shelf.

[0110] Figure 11(c) shows the Z-axis positioning error of the algorithm proposed in this invention and the MRL algorithm as a function of the target length in the shelf.

[0111] Figure 12 This is a CDF plot of the 3D positioning error. Detailed Implementation

[0112] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0113] This invention proposes a Mobile Dual-Tag Dual-Antenna RFID Positioning System (MRRDT). First, a robot collects the RSS and phase information of the reflected signals from tags affixed to a target object during its movement. Then, the peak RSS signal is used to initially estimate the x-dimensional value of the target, and a dual-tag phase model is used to estimate the y and z dimensions of the target. Subsequently, a three-dimensional MRL is used to calibrate the coarse estimate of the target's position, thereby obtaining the target's three-dimensional position information.

[0114] (I) Implementation process of three-dimensional target perception method

[0115] Taking a shelf model as an example, this invention addresses the problem of sensing the location information of items placed on a shelf. The RFID positioning system architecture constructed in this embodiment is as follows: Figure 1 As shown, each item on the shelf has two stickers with a spacing of d. a,b Equal passive RFID tags (T a T b(Since these two tags appear in pairs, they will be referred to as tag pairs below.) In addition, a mobile robot was placed on a horizontal surface, on which were mounted a portable computer and an RFID reader (the reader is equipped with two directional antennas, denoted as A1 and A2 respectively, directly connected to the reader, with both antennas facing the shelf, vertically placed on the robot platform, and the center-to-center distance between the antennas is d). 1,2 The distance between the lower antenna A2 and the ground is h. ant ).

[0116] First, a mobile robot moves from one end of a shelf to the other at a fixed speed *v*. Second, a reader carried by the robot sends a signal via an antenna to activate passive tags affixed to objects on the shelf, which then backscatter. The reader's dual antennas collect the backscattered signals from the tags. Finally, a portable computer stores and processes the collected tag backscattered signals and uses the three-dimensional high-precision target perception algorithm of this invention to estimate the three-dimensional position information of the goods on the shelf.

[0117] Let the origin O be the projection of the robot's center of gravity onto the ground at the initial moment of its motion. Let the robot's direction of motion be the x-axis, its direction perpendicular to the ground be the z-axis, and the y-axis be perpendicular to the plane formed by xoz. According to... Figure 1 As defined in [the original text], the starting coordinates of the upper and lower antennas of the reader can be represented as (0, 0, h) respectively. ant +d 1,2 ) and (0,0,h ant ).

[0118] The three-dimensional high-precision target perception algorithm of this invention is divided into two stages: coarse estimation of the target's three-dimensional position and position calibration. The specific process is as follows: Figure 2 As shown, firstly, a mobile robot equipped with an RFID reader is used to perceive the phase information of all tagged targets on the shelf using a single reader, thereby sensing the height and depth information of the targets relative to the shelf. Then, signal strength information, combined with a path loss model, is used to perceive the column position of the targets relative to the shelf. Finally, based on RSS peak estimation, the x-dimensional estimate is obtained, and the initial y and z-dimensional estimates are obtained using a dual-tag phase model. Using the positions estimated in these three dimensions as initial values, the target position is further corrected based on the MRRDT algorithm proposed in this invention, thus obtaining the three-dimensional high-precision position information of the targets on the shelf. The specific implementation steps are as follows:

[0119] S1, RFID reader is mounted on a robot, and the robot's movement collects the RSS and phase information reflected by a set of tags affixed to the target being sensed; and the RSS and phase information are filtered.

[0120] The total time period of the robot's movement is sampled N times, and the label is T. a In t n The RSS value obtained at time (n∈[1,N]) from the i-th antenna (i=1,2 correspond to the upper and lower antennas of the reader, respectively) is rss. a,i (t n ), tag T b In t n The RSS value obtained from the i-th antenna at time t is rss b,i (t n Then, the total motion time of the robot can be obtained by collecting all the RSS values, which can be formed into a matrix as follows:

[0121] RSS = [rss] a,1 rss b,1 rss a,2 rss b,2 ] T (1)

[0122] in,

[0123] rss a,i =[rss a,i (t1),...,rss a,i (t n ),...,rss a,i (t N )]

[0124] rss b,i =[rss b,i (t1),...,rss b,i (t n ),...,rss b,i (t N )]

[0125] Since each tag has a unique RSS profile curve, this uniqueness can be used to distinguish different tags. Assume that in t... n The signal strength received by the time reader antenna i from tag a is rss. a,i (t n The signal strength from tag b is rss. b,i (t n Then we have:

[0126]

[0127]

[0128] Where S represents the number of samples, t n,s Indicates at time t nThe RSS obtained from the s-th sampling point.

[0129] Define the signal strength difference between tag a and tag b as disRSS. a,b Then we have:

[0130]

[0131] After the RFID reader collects a set of RSS and phase information, due to measurement errors and noise, the RSS and phase information are first filtered separately, and then the coarse position of the target is estimated based on the filtered information.

[0132] In real-world environments, even when both the RFID reader and the passive tag are stationary, the RSS value received by the reader from the same tag at different times is not constant. Noise or multipath effects in the communication environment can cause this. (The last sentence appears to be a separate, unrelated statement: "The reader's upper antenna receives the RSS value of the tag T from the target at a fixed point.") a Taking the RSS signal as an example, such as Figure 3 As shown, the initial signal fluctuates significantly, with an average RSS value of -41dBm, a maximum value of -37dBm, and a minimum value of -47dBm, with a difference of 10dBm between the maximum and minimum values.

[0133] Therefore, when there is limited sampled data, directly using the average of these values ​​may not be accurate. Thus, in such cases, it is necessary to filter the RSS measurement signal. First, abnormal RSS measurements are removed, and then the average of other RSS values ​​collected at the same time is used as the RSS value received by the antenna from the tag under test at that time.

[0134] Figure 3 The paper also presents the effects of Gaussian filtering, mean filtering, and median filtering on removing outliers from RSS measurements. Figure 3 It can be observed that the signal obtained by Gaussian filtering to remove outliers has good continuity and retains more details of the original signal. Therefore, Gaussian filtering is used to filter the original RSS measurements in this embodiment.

[0135] First, consider the time t in the RSS matrix shown in equation (1). n The acquired signal strength value rss(t) n Gaussian filtering is performed, and then the average of multiple sets of values ​​obtained at the same time is taken to complete the smoothing process. The processed RSS matrix can be represented as:

[0136] RSS′=[rss′ a,1 rss′ b,1 rss′ a,2 rss′ b,2] T (5)

[0137] in,

[0138] rss′ a,i =[rss' a,i (t1),...,rss′ a,i (t n ),...,rss′ a,i (t N )]

[0139] rss′ b,i =[rss' b,i (t1),...,rss′ b,i (t n ),...,rss′ b,i (t N )]

[0140] Where i = 1, 2.

[0141] S2. Obtain the peak time of the tag's RSS based on the filtered RSS information, multiply this peak value by the robot's speed to obtain the tag's approximate position information relative to the shelf length direction; then, estimate the approximate position information of the perceived target along the shelf depth and height directions based on the preprocessed phase information.

[0142] Step S21: Estimate the coarse x-dimensional position of the target based on the RSS measurement.

[0143] From the RSS signal obtained after preprocessing in step S1, it can be observed that the closer the antenna is to the tag, the stronger the signal. Therefore, the x-dimensional coordinates of the tag under test can be estimated by finding the peak value of the processed RSS signal. For example, three tags are placed at positions (0.45, 0.6, 0.7), (1.56, 0.9, 1), and (2.83, 0.8, 0.5) respectively. The unit is m. The RSS signal strength values ​​after preprocessing are as follows: Figure 4 As shown.

[0144] Depend on Figure 4It can be seen that for these three tags, the peak values ​​of the RSS signals acquired by the upper antenna A1 appear at 3s, 8s, and 14s, respectively; the peak values ​​of the RSS signals acquired by the lower antenna A2 appear at 3s, 7s, and 15s, respectively. In this embodiment, v = 0.2m / s is used. Therefore, the x-coordinates corresponding to the three tags acquired by the upper antenna A1 are 0.6m, 1.6m, and 2.8m, respectively; and the x-coordinates corresponding to the three tags acquired by the lower antenna A2 are 0.6m, 1.4m, and 3m, respectively. In summary, averaging the above two sets of results, the x-coordinates corresponding to the peak values ​​of the RSS signals for the three tags are 0.6m, 1.5m, and 2.9m, respectively.

[0145] Therefore, let the times corresponding to the peak values ​​of RSS′ in equation (5) be t. a,1 t a,2 t b,1 t b,2 Let t ⊥ Let the moment when the target is closest to the antenna be represented by:

[0146]

[0147] Therefore, the initial position estimate of the target in the x-dimensional dimension is... It can be represented as:

[0148]

[0149] Where v represents the robot's moving speed.

[0150] Step S22: Estimate the target's y and z dimension positions based on the phase measurement values.

[0151] During robot movement, the phase matrix of the backscattered signal from the tag pair of the target to be perceived can be acquired.

[0152]

[0153] in,

[0154]

[0155]

[0156] i = 1, 2 correspond to the upper and lower antennas of the reader, respectively. However, due to multipath effects, environmental noise, and phase deflection caused by differences in equipment and tags, the phase of the tag backscattered signal received by the antenna is interfered with. Let δ represent the phase deflection caused by the environment or equipment. Let tag T... a For example, the reader's antenna i in t n Phase acquired at each moment It can be represented as:

[0157]

[0158] Where, d a,i (t n ) represents t n Time Antenna A i With label T a The distance between them, where k is a non-negative integer (due to the periodicity of the phase).

[0159] Since there are no jumps between consecutive phase values, outliers can be identified and compensated for by checking for phase jumps between adjacent time points, thus ensuring continuity between adjacent phase values. The compensated phase... It can be represented as:

[0160]

[0161] Phase compensation is performed on the N sets of phases collected during the total motion time of the robot to obtain the phase matrix:

[0162]

[0163] in,

[0164]

[0165]

[0166] i = 1, 2 correspond to the upper and lower antennas of the reader, respectively.

[0167] Besides the phase change caused by distance variation, the phase deflection δ caused by other factors was not eliminated. Therefore, in this embodiment, phase value calibration is performed after phase unfolding. A tag with a known location is selected as a reference tag, and a mobile robot with a single step size of 5 cm is used to measure the phase of the tag's backscattered signal, with each measurement lasting 30 seconds. The phase measurement values ​​are then... Phase expansion yields Take the theoretical value of the phase Phase value after phase expansion The difference C between them is used as the phase correction matrix, and the expression for the difference C is as follows:

[0168]

[0169] Based on the correction matrix C shown in equation (12), equation (11) can be further modified to obtain the corrected phase matrix Φ. u′ ,

[0170] Φ u′=Φ u -C (13)

[0171] in,

[0172]

[0173] in,

[0174]

[0175]

[0176] i = 1, 2 correspond to the upper and lower antennas of the reader, respectively.

[0177] After completing the phase matrix preprocessing, since it is difficult to directly estimate the position of the target in the y and z dimensions using a single label, this embodiment introduces a dual-label phase model algorithm to complete the position estimation in these two dimensions. The dual-label phase model is as follows: Figure 5 As shown.

[0178] Depend on Figure 5 It can be seen that the target to be tested has two labels, upper and lower, affixed respectively. a and T b The distance between the two labels is d a,b d a,i (t n ) and d b,i (t n ) represent time t respectively n The i-th antenna of the reader and the tag T a and T b The distance between them,

[0179]

[0180] in, They represent t respectively n Moment tag T a T b The phase calibration value between antenna A and the i-th antenna of the reader, where λ represents the wavelength of the electromagnetic wave emitted by the reader. Therefore, antenna A i At every moment of the entire process with label T a T b The corresponding distance vector d a,i d b,i They can be represented as:

[0181] d a,i =[d a,i (t1),...,d a,i (t n ),...,d a,i (tN (15)

[0182] d b,i =[d b,i (t1),...,d b,i (t n ),...,d b,i (t N (16)

[0183] exist Figure 5 In the middle, in t n At time, A i The location is A i (t n ), passing through point A i (t n ) as label T a T b The perpendicular line to the line intersects at point G. i So, label T a and A i (t n ) and G i The three points can form a right triangle, and the label T b and A i (t n ) and G i The three points can also form a right triangle, therefore at time t n have

[0184]

[0185] in, Indicates t n Moment tag T b and projection point G i The distance between them. According to equation (17), it can be calculated as follows:

[0186]

[0187] Then according to Figure 5 By considering the positional relationship of the target to be measured, the estimated z-dimensional position of the target at this moment can be obtained. T,i (t n The expression that satisfies this condition is as follows:

[0188]

[0189] Solving equation (18) yields Substituting into equation (19) yields z T,i (t n Then, estimate the z-dimensional value of the target at all times. T,i Taking the average, we have

[0190]

[0191] At the same time, in label T b and A i (t n ) and G i The right triangle formed by the three points contains

[0192]

[0193] In equation (21) For t n Point G at time i With antenna A i (t n The distance between them can be solved by substituting equation (18) into equation (18). However, the true position in the y-dimensional dimension is only when the antenna is perpendicular to the label t. ⊥ We need to find the value of y at each step, so that the actual value of y is... Now, let's solve for t by combining the entire process. n The y-value corresponding to time t is:

[0194]

[0195] In the formula t ⊥ It can be obtained through equation (6), where v represents the robot's moving speed. Then, consider y(t) in equation (22). n ) Calculate the average value of the entire process. Among them, t n ∈[t1,t N ], then we have:

[0196]

[0197] The dual-label phase model is described using a single antenna, but to obtain more accurate results, the two sets of data obtained from antennas A1 and A2 are compared. (i = 1, 2 correspond to the upper and lower antennas respectively) Taking the average value, we have:

[0198]

[0199] Combining equations (7) and (24), the initial rough estimate of the target's position is:

[0200] Step S2, x-dimensional position calibration

[0201] Due to interference and noise during signal measurement, the initial position estimation based on RSS signal measurements has a large error. This invention proposes an x-dimensional calibration algorithm based on Taylor series expansion to further calibrate the x-dimensional initial position shown in equation (5). A schematic diagram of the x-dimensional calibration algorithm is shown below. Figure 6 As shown.

[0202] Depend on Figure 6 It can be seen that the entire time process t N It is divided into three equal parts, namely [t1,...,t] w ]、[t w+1 ,...,t 2w ]、[t 2w+1 ,...,t N ], where N = 3w. Let t n t n+w and t n+2w The antenna A1 positions at the three time points are R1, E1, and J1, respectively, and the antenna A2 positions at the same time points are R2, E2, and J2, respectively. A calibration algorithm is then used to calibrate the coarse 3D estimate of the target object. This algorithm requires at least one tag and two antennas from the reader to participate. Let tag T... a For example, label T a The four points R1, E1, and J1 form a plane, as shown below. Figure 7 As shown.

[0203] like Figure 7 As shown, let label T a The projection point of the line containing points R1, E1, and J1 is M1. Therefore, points R1, E1, and J1 are respectively aligned with label T. a Points M1 and M2 form three right triangles. and Therefore:

[0204]

[0205] Then from label T a There is a corresponding relationship between the phase and distance of backscattering at different times, which can be obtained as follows:

[0206]

[0207] In X a,1 A1(t1)Y a,1 In a plane, label T a The position can be represented as (x a,1 ,y a,1 ), combined with the label T a exist Figure 6 and Figure 7The positional relationships in the coordinate system can be obtained as follows:

[0208]

[0209] Furthermore, due to the velocity v and the time point t n t n+w t n+2w Given that, we can obtain

[0210]

[0211] Combining equations (25) to (28), we can obtain:

[0212]

[0213] Since directly solving the nonlinear equations shown in equation (29) is quite difficult, a Taylor series expansion is performed on equation (29), and the rough estimates of the x-dimensional and y-dimensional dimensions of the target obtained from equations (7) and (24) are taken. As an initial value, we can obtain:

[0214]

[0215] in,

[0216]

[0217] Thus, the label T can be obtained from formula (30). a X in this plane a,1 Axis, Y a,1 The projections of the axes are x a,1 y a,1 , respectively corresponding for Figure 6 , The label T for the target to be tested a The true position of the x-axis, but Let M1 be the distance from antenna A1 to projection point M1.

[0218] With acquisition Similarly, label T a The plane formed by points R2, E2, and J2 can be calculated. in Now, the x values ​​obtained from the two antennas are... a,1 x a,2 For the estimated label T a At the x-axis position, taking the average value yields:

[0219]

[0220] in, x aFor label T a The true x-axis position estimate.

[0221] Let the lines containing M1 and M2 intersect at point o on the x-axis. M Let the line containing M1 and M2 be z. M Axis, through o M The line parallel to the y-axis is y M Axis. Therefore, y M o M z M Plane as follows Figure 8 As shown.

[0222] like Figure 8 As shown, label T a At this time in z M The axis projection points are denoted as p, M2 and label T. a Forming a right triangle and p, M1 and label T a They form a right triangle, therefore we can conclude:

[0223]

[0224] in The distance between point p and point M1 can be obtained from equation (32):

[0225]

[0226] Then, according to equation (33), Figure 8 The existing positional relationships can be obtained

[0227]

[0228] Combining equations (31) and (34), the estimated tag T based on the calibration method proposed in this invention can be obtained. a 3D coordinates (x) a ,y a ,z a ).

[0229] The above obtained (x) a ,y a ,z a ) for t n t n+w t n+2w Labels T obtained through calibration algorithms at three time points a The three-dimensional coordinates are determined. Since the time length is N = 3w, w possible coordinate positions can be solved during the three-dimensional position calibration stage.

[0230] Simply averaging these w positions usually doesn't yield optimal results. Therefore, weights are introduced to further optimize the algorithm's localization performance. The reader's movement direction is taken as the inverse of the variance of the estimated values ​​in the x-dimensional region. Using these weights to further calibrate the x-dimensional coordinates, we obtain:

[0231]

[0232] Similar to equations (25)-(35) above, the updated label T b The x-dimensional coordinates are x b,fin Therefore, the x-coordinate of the target object can be updated as follows:

[0233]

[0234] Combining equations (24) and (36), the final three-dimensional coordinate position of the target to be measured can be estimated as follows:

[0235] (II) Experimental Verification

[0236] To verify the performance of the MRRDT algorithm proposed in this invention, tests were conducted within a 4m × 4m × 2.5m area. The shelf was 2m long, 2m high, and 1m deep. The target objects to be tested were arranged on the shelf as follows: Figure 1 As shown, 18 objects are arranged in three rows and six columns. Each object has two tags, one above and one below, for a total of 18 tag pairs. The tag pairs are spaced 5 cm apart. The tags include five types: Alien-964X, Impinj E41B, E41C, H47, and AZ-E53.

[0237] The selected mobile robot was a Pioneer 3-DX, with a default speed of 0.2 m / s. The reader was an Impinj Speedway R420, set to a frequency of 924.5 MHz. Two sets of directional antennas were LAIED s9025 antennas, with a transmit power of 32.5 dBm, an antenna gain of 9 dBic, and a sampling rate of 10 Sa / s. The antennas were positioned at heights of 0.3 m and 0.6 m above the ground, respectively. The portable computer mounted on the mobile robot had an AMD R7 5800H CPU and 16 GB of RAM. The experimental equipment is shown in Table 1.

[0238] Table 1. Description of Experimental Equipment

[0239]

[0240] To evaluate the positioning performance of the algorithm of this invention, the root mean square error (RMSE) of positioning is used for evaluation:

[0241]

[0242] Wherein, the P sequence represents the true positions of the m targets to be measured in this system, and P = [P1, P2, ..., P...]. m ]; The sequence represents the estimated locations of m targets to be measured in this system, and

[0243] (1) Target perception accuracy analysis

[0244] Based on the above settings, the targets to be measured were divided into three groups, and their depth (keeping the length and height directions constant), height (keeping the length and depth directions constant), and length (keeping the depth and height directions constant) were changed respectively. The results are shown in Figures 9 to 11. For a group of targets with fixed length and height, the coordinates were set as follows: Target1 (1.3, 0.6, 1.3), Target2 (1.3, 0.8, 1.3), and Target3 (1.3, 1, 1.3). For this group of targets, Figure 9 shows the relationship between the positioning error corresponding to the two different algorithms and the change of the target in the shelf depth direction (y-axis).

[0245] As shown in Figure 9, for the x-dimensional position of the target, the root mean square error (RMSE) of both the MRL method and the method proposed in this invention (MRRDTfinal, Error in x) is controlled below 1 cm, while the initial position coarse estimation (MRRDTfirst, Error in x) of this invention is controlled at around 5 cm. Secondly, for the y and z dimensions, the results obtained by the dual-label phase estimation (MRRDTfirst, Error in y or z) of this invention also reduce the RMS error to below 1 cm. Finally, regarding the positioning performance of the MRL algorithm and similar calibration algorithms for the x-dimensional position, the positioning error increases as the target is placed further back in the shelf depth direction, while the results obtained by the dual-label phase model of this invention are less affected by this.

[0246] Among them, the target group for which the robot's movement direction and its depth position on the shelf only change in height are: Target4 (1.3, 0.8, 0.7), Target5 (1.3, 0.8, 1.3), and Target6 (1.3, 0.8, 1.9). For this group of targets, Figure 10 shows the relationship between the positioning error corresponding to the two different algorithms and the change of the target's height in the shelf direction, i.e., the z-axis.

[0247] As shown in Figure 10, for the x-dimensional target, the root mean square error of the results calibrated by the MRL method and the proposed method (MRRDTfinal, Error in x) is controlled below 1 cm. Secondly, for the y and z dimensions, the results obtained by the dual-label phase estimation (MRRDTfirst, Error in y or z) of the proposed method also reduce the root mean square error to below 1 cm. Finally, as the height of the target increases, the localization performance of other methods becomes relatively worse in other dimensions, except for the stable localization performance of the dual-label phase model in the z-dimensional.

[0248] Among them, a set of targets to be measured, fixed at the depth and height of the shelf, are Target7 (0.7, 0.8, 1.3), Target8 (1.3, 0.8, 1.3), and Target9 (1.9, 0.8, 1.3). For this set of targets, Figure 11 shows the relationship between the positioning error corresponding to the two different algorithms and the change of the target in the x-axis direction of the reader's movement.

[0249] As shown in Figure 11, for the x-dimensional target, the root mean square error (RMSE) obtained by the MRL method and the calibration algorithm of this invention (MRRDTfinal) is controlled below 1 cm. Secondly, for the y and z dimensions, the dual-label phase estimation (MRRDTfinal) of this invention also reduces the RMSE to below 1 cm. However, using the initial value obtained from the dual-label phase estimation for a similar x-dimensional calibration algorithm estimation results in significantly lower performance than the positioning results obtained before calibration. Based on these results, this invention uses MRDTfinal for x-dimensional estimation and MRDTfirst for y and z dimensions, i.e., employing... This serves as the final output of the shelf positioning system. It's worth noting that if the robot doesn't reach the RSS peak during its movement, the 3D position obtained from the position calibration value of this invention is more accurate. However, once the robot reaches the target label's RSS peak, the accuracy of the position obtained from the coarse estimation of the y and z dimensions improves. The average error of the target estimation using the MRL method and the calibration algorithm of this invention is shown in Table 2.

[0250] Table 2. Average error of the two algorithms (cm)

[0251]

[0252] (2) Target perception performance

[0253] Figure 12 The cumulative distribution function (CDF) curves of the three-dimensional positioning error of the shelf target using the method of this invention and the MRL algorithm were plotted. From... Figure 12As can be seen, the average errors of this invention on the x-axis, y-axis, and z-axis are 0.69cm, 1.02cm, and 0.05cm, respectively, and the average three-dimensional error of the shelving system is 1.34cm. The z-axis positioning performance is the best, followed by the x-axis and y-axis positioning performance. The MRL algorithm performs similarly to the algorithm of this invention in the x-axis dimension, and even slightly better in some areas; however, its positioning performance in the y-axis, z-axis, and even the overall positioning performance of the target is inferior to the algorithm proposed in this invention.

[0254] Simulation results show that the MRRDT system proposed in this invention can achieve centimeter-level positioning accuracy while effectively reducing deployment costs and improving equipment utilization.

Claims

1. A three-dimensional high-precision target perception method based on a mobile RFID reader and dual tags, characterized in that, The steps include the following: S1, The RFID reader is mounted on the robot, and the robot's movement collects the RSS and phase information reflected by a set of tags pasted on the target being sensed; and the RSS and phase information are filtered. S2. Obtain the peak time of the tag's RSS based on the filtered RSS information, and multiply this peak time by the robot's speed to obtain the approximate position information of the tag relative to the x-axis of the shelf length direction. Subsequently, based on the preprocessed phase information, the approximate position information of the perceived target along the shelf depth direction (y-dimensional) and height direction (z-dimensional) is estimated. S3 uses the obtained length, height, and depth information of the target to be measured as initial values, and further calibrates the position information of the perceived target based on the MRRDT algorithm to obtain accurate position information.

2. The three-dimensional high-precision target perception method based on a mobile RFID reader and dual tags according to claim 1, characterized in that, In step S1, the total time period of the robot's movement is sampled N times, and the label T is set to... a In t n The RSS value obtained from the i-th antenna at time t is rss a,i (t n ), tag T b In t n The RSS value obtained from the i-th antenna at time t is rss b,i (t n ); where i = 1, 2, corresponding to the upper and lower antennas of the reader, respectively; n ∈ [1, N], then all RSS values ​​collected during the total motion time of the robot are obtained, forming a matrix as follows: RSS=[rss a,1 RSS b,1 RSS a,2 RSS b,2 ] T in, rss a,i =[rss a,i (t1),...,rss a,i (t n ),...,rss a,i (t N )] rss b,i =[rss b,i (t1),...,rss b,i (t n ),...,rss b,i (t N )] Where S represents the number of samples, t n,s Indicates at time t n The RSS obtained from the s-th sampling point; Define the signal strength difference between tag a and tag b as disRSS. a,b Then we have: Among them, rss a (t n ) for in t n The time reader receives the signal strength from tag a, rss b (t n ) for in t n The time reader receives the signal strength from tag b.

3. The three-dimensional high-precision target perception method based on a mobile RFID reader and dual tags according to claim 2, characterized in that, In step S1, for time t n The acquired signal strength value rss(t) n Gaussian filtering is performed, and then the average of multiple sets of values ​​obtained at the same time is taken for smoothing. The resulting RSS matrix is ​​RSS′, which is represented as follows: RSS′=[rss′ a,1 rss′ b,1 rss′ a,2 rss′ b,2 ] T in, rss′ a,i =[rss′ a,i (t1),...,rss′ a,i (t n ),...,rss′ a,i (t N )] rss′ b,i =[rss′ b,i (t1),...,rss′ b,i (t n ),...,rss′ b,i (t N )] Where i = 1 and 2, corresponding to the upper and lower antennas of the reader, respectively.

4. The three-dimensional high-precision target perception method based on a mobile RFID reader and dual tags according to claim 3, characterized in that, In step S2, the initial position estimation of the target in the x-dimensional dimension is achieved as follows: Let the times corresponding to the peak values ​​of RSS′ be t. a,1 t a,2 t b,1 t b,2 , let t ⊥ Let the moment when the target is closest to the antenna be represented by: Obtain the initial position estimate of the target in the x-dimensional dimension. Represented as: Where v represents the robot's moving speed.

5. The three-dimensional high-precision target perception method based on a mobile RFID reader and dual tags according to claim 3, characterized in that, In step S2, a dual-label phase model is used to estimate the position of the target in the y and z dimensions. The specific implementation steps are as follows: S21, the target to be tested is labeled with two tags, T, which are attached to the top and bottom respectively. a and T b The distance between the two labels is d a,b d a,i (t n ),d b,i (t n ) represent time t respectively n The i-th antenna of the reader is related to the tag T a and T b The distance between them is: in, They represent t respectively n Moment tag T a T b The phase calibration value between the i-th antenna of the reader and the reader, where λ represents the wavelength of the electromagnetic wave emitted by the reader; S22, Antenna A i At every moment of the entire process with label T a T b The corresponding distance vector d a,i d b,i They are represented as follows: d a,i =[d a,i (t1),...,d a,i (t n ),...,d a,i (t N )] d b,i =[d b,i (t1),...,d b,i (t n ),...,d b,i (t N )] In t n At time, A i The location is A i (t n ), passing through point A i (t n ) as label T a T b The perpendicular line to the line intersects at point G. i So, label T a and A i (t n ) and G i The three points can form a right triangle, and the label T b and A i (t n ) and G i The three points can also form a right triangle; therefore, at time t n have: in, Indicates t n Moment tag T b and projection point G i The distance between them; Then we have: The positional relationship of the target under test is used to estimate the target's z-dimensional position at this moment. T,i (t n The following equation must be satisfied: Then estimate the z-dimensional value of the target at all times. T,i Taking the average, we have: At the same time, in label T b and A i (t n ) and G i The right triangle formed by the three points contains: in, For t n Point G at time i With antenna A i (t n The distance between them; However, the true position in the y-dimensional dimension is only when the antenna is perpendicular to the label t. ⊥ We obtain the value at each step, and let the actual value of y be... Solving t in conjunction with the entire process n The y-value corresponding to time t is: In the formula t ⊥ This indicates the moment when the target is closest to the antenna, and v represents the robot's moving speed; Then for y(t) n ) Calculate the average value of the entire process. Among them, t n ∈[t1,t N ], then we have: The two sets of data obtained from the upper antenna A1 and the lower antenna A2 Taking the average, we have: Finally, the initial coarse estimate of the target's position is obtained. This is the initial position estimate of the target in the x-dimensional dimension.

6. The three-dimensional high-precision target perception method based on a mobile RFID reader and dual tags according to claim 5, characterized in that, In step S3, a calibration algorithm based on Taylor series expansion is used to adjust the initial position estimate. Further calibration, specifically implemented through the following steps: The entire time process t N It is divided into three equal parts, namely [t1,...,t] w ]、[t w+1 ,...,t 2w ]、[t 2w+1 ,...,t N ], where N = 3w; let t n t n+w and t n+2w The positions of the upper antenna A1 at the three time points are R1, E1, and J1, respectively, and the positions of the upper antenna A2 at the same time points are R2, E2, and J2, respectively; tag T a Points R1, E1, and J1 form a plane. Let label T. a The projection point of the line containing points R1, E1, and J1 is M1. Therefore, points R1, E1, and J1 are respectively aligned with label T. a Points M1 and M2 form three right triangles. and Therefore: Then from label T a There is a correlation between the phase and distance of backscattering at different times, which yields: In X a,1 A1(t1)Y a,1 In a plane, label T a The position is represented as (x a,1 ,y a,1 ), combined with the label T a The positional relationship between the three-dimensional calibration coordinate system and the two-dimensional planar coordinate system is obtained as follows: Due to velocity v and time point t n t n+w t n+2w Given that, we have: Combining the above four equations, we get: A Taylor series expansion is performed, and coarse estimates are made of the x and y dimensions of the target object. As an initial value, we get: in, Get label T a X in a two-dimensional plane a,1 Axis, Y a,1 The projections of the axes are x a,1 y a,1 , respectively corresponding in, The label T for the target to be tested a The true position of the x-axis; Similarly, label T a The plane formed by points R2, E2, and J2 is obtained. in, The label T for the target to be tested a The true position of the x-axis; The obtained x a,1 x a,2 For the estimated label T a Taking the average value at the x-axis position, we have: Where, x a For label T a Estimate the true x-axis position; Let the lines containing M1 and M2 intersect at point o on the x-axis. M Let the line containing M1 and M2 be z. M Axis, through o M The line parallel to the y-axis is y M Axis; in the yoz plane coordinate system of the calibration algorithm, label T a At this time in z M The axis projection points are denoted as p, M2 and label T. a Forming a right triangle and p, M1 and label T a Forming a right triangle, we get: in, Let p be the distance between points M1; finally, we get: Then the three-dimensional coordinates (x a ,y a ,z a ) for t n t n+w t n+2w Labels T obtained through calibration algorithms at three time points a 3D coordinates; Take the reciprocal of the variance of the estimated values ​​in the x-dimension. The x-dimensional coordinates are calibrated using weights, resulting in the updated x-dimensional coordinates of the target object: Where, x a,fin For the updated tag T a x-dimensional coordinates, x b,fin For the updated tag T b x-dimensional coordinates; Finally, the estimated final three-dimensional coordinate position of the target is obtained as follows:

Images