UWB-based shopping guide system and method for shopping mall

Abstract

The invention discloses a UWB-based market shopping guide system and method, and relates to the field of intelligent shopping guide, and the system comprises a positioning module, an acquisition fusion module, a model construction module, a shopping guide path generation module, an interference compensation construction module and a credibility evaluation module. According to the method, the problem of insufficient precision of WiFi and Bluetooth methods is effectively solved by combining a time difference arrival and arrival angle combined positioning mode of UWB. A five-level space mapping index is established, direct correspondence between a customer target and an actual position is achieved, and the commodity retrieval and navigation accuracy of the shopping guide system is improved. The machine learning model is used for modeling the residence time, the access frequency and the promotion interaction behaviors of the customers, and the preference distribution of the customers is dynamically calculated, so that personalized recommendation and interest area navigation are provided in the shopping guide process. A congestion prediction model and a customer behavior prediction mechanism are introduced in a path planning process, a high-flow area is dynamically avoided, an optimal shopping path is generated, and shopping guide efficiency and comfort are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent shopping guide technology, and in particular to a shopping mall shopping guide system and method based on UWB. Background Technology

[0002] As modern commercial complexes and shopping malls continue to expand, traditional shopping guide methods are no longer sufficient to meet customers' real-time navigation and personalized needs. Especially in large shopping malls, where stores are complexly distributed, there are multiple floors, and customer traffic is high, customers often find it difficult to find their target products or quickly choose the optimal shopping route.

[0003] Existing indoor positioning and shopping guide technologies mainly fall into the following categories: GPS has high accuracy in outdoor environments, but due to building obstruction, signal attenuation, and multipath effects, it is almost unusable indoors, thus failing to meet the navigation needs of shopping malls. Positioning via WiFi fingerprints or Bluetooth beacons can achieve some degree of navigation within the mall, but its accuracy is typically within the 1-3 meter range, significantly affected by signal interference, device power consumption, and deployment costs, making it impossible to achieve refined navigation at the shelf or product level. Most shopping guide systems only provide store-level navigation, lacking the ability to locate specific shelves or products; they lack real-time integration with the mall's product database and promotional information, failing to achieve customer interest-driven intelligent recommendations; existing path planning is mostly based on static shortest paths, without considering customer flow distribution, environmental congestion, and customer shopping habits, resulting in insufficient practicality of navigation results; the reliability of navigation positioning data lacks effective quantification and dynamic feedback mechanisms, and when positioning data is distorted, it can easily lead to shopping guide failures or even mislead customers.

[0004] In summary, existing technologies have significant shortcomings in indoor high-precision positioning, intelligent shopping guides, dynamic path optimization, and credibility assessment. There is an urgent need for a shopping mall shopping guide method that combines UWB centimeter-level positioning, machine learning preference modeling, path optimization, and dynamic credibility scoring to improve the accuracy, reliability, and personalized service level of shopping guides. Summary of the Invention

[0005] In view of the problems of insufficient positioning accuracy, lack of personalized shopping guides, unreasonable route planning, and inability to guarantee the reliability of positioning data, this invention is proposed.

[0006] Therefore, the problem to be solved by this invention is how to achieve intelligent identification of customer interests and preferences, generate dynamic optimal shopping guide paths, and establish a navigation credibility evaluation mechanism.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] In a first aspect, embodiments of the present invention provide a shopping mall guide system based on UWB, including: a positioning module, used to deploy UWB base stations on each floor and passage of the shopping mall, and to achieve centimeter-level precision positioning of customer terminals through a combination of time difference of arrival and angle of arrival positioning method;

[0009] The data collection and fusion module is used to collect customers' movement speed, dwell time and historical trajectory, and combine it with data from the mall's WiFi and Bluetooth beacons for multi-source fusion. Through machine learning models, it identifies customers' shopping preferences and areas of interest in real time.

[0010] The model building module is used to establish a spatial mapping relationship between product location and UWB coordinate system based on the 3D shopping mall building information model and product database. It adopts a hierarchical partition index method to divide the shopping mall into five levels of mapping: floor, area, store, shelf, and product.

[0011] The shopping route generation module is used to dynamically generate the optimal shopping route that avoids high-traffic areas by using route planning algorithms and congestion prediction models. It also incorporates customer behavior prediction models to optimize the route based on shopping order and dwell habits, and adds weight constraints of environmental factors during the route generation process.

[0012] The interference compensation module is used to dynamically correct the positioning data using a joint compensation algorithm of Kalman filtering and particle filtering when there is electromagnetic interference or signal blockage in the shopping mall. At the same time, it performs self-learning of positioning error patterns based on deep neural networks.

[0013] The credibility assessment module is used to establish a navigation credibility index, calculate the confidence level of the compensated location data, dynamically generate a credibility score, and automatically trigger alternative route planning or multi-source data weighted fusion when the credibility level is lower than the threshold, so as to realize the switching and protection of the shopping guide route.

[0014] As a preferred embodiment of the UWB-based shopping mall navigation system of the present invention, the system achieves centimeter-level precision positioning of customer terminals through a combined positioning method of time difference of arrival and angle of arrival, including:

[0015] The customer terminal transmits a UWB pulse, and the base station collects the channel impulse response and performs high-resolution reconstruction of the channel impulse response using sparse reconstruction or multi-resolution time-domain denoising algorithms. The arrival time of the first path is identified using an adaptive threshold or multi-peak peak separation method based on the energy rising edge, and the corresponding time difference arrival observation is calculated. At the same time, the original angle of arrival measurement is extracted from the array phase difference or covariance matrix.

[0016] As a preferred embodiment of the UWB-based shopping mall guide system described in this invention, the system includes: real-time identification of customers' shopping preferences and areas of interest through a machine learning model, including:

[0017] Customers are grouped using a clustering algorithm. The clustering algorithm function is expressed as follows:

[0018]

[0019] Where, μ k For the center of class k, F i Let be the feature vector of the i-th customer;

[0020] A deep time-series prediction model is used to predict the next area of ​​customer interest. The function of the deep time-series prediction model is:

[0021] P(a t+1 |F t =Softmax(W·h) t +b)

[0022] Among them, h t For the temporally hidden state, P(a t+1 For customers entering area a in the next time slot t+1 The probability distribution;

[0023] Based on real-time input features F t Dynamically calculate the distribution of customer preferences:

[0024]

[0025] Where α, β, γ are weight parameters, Pref j Let Δt represent the real-time probability of a customer's preference for the j-th region. j f is the time the customer spends in the j-th area. j r represents the number of times a customer revisits in the j-th region. j Let Δt be the frequency of customer interaction with promotional information in the j-th area, m be the total number of areas within the mall, and Δt be the frequency of interaction with promotional information in the j-th area. k f is the time a customer spends in the kth region. k r is the number of times a customer revisits in the k-th region. k The frequency of customer interaction with promotional information in the k-th region.

[0026] As a preferred embodiment of the UWB-based shopping mall navigation system described in this invention, the system establishes a spatial mapping relationship between product locations and the UWB coordinate system based on a 3D shopping mall building information model and a product database. A hierarchical partitioning index is used to divide the shopping mall into five levels: floor, area, shop, shelf, and product.

[0027] Map the points (x, y, z) in the UWB positioning coordinate system to the BIM spatial coordinate system using the spatial affine transformation matrix M:

[0028] (x′,y′,z′)=M·(x,y,z,1) T

[0029] Where (x′,y′,z′) represents the customer's corresponding location in the BIM space;

[0030] A five-level index structure is used to spatially map customer locations:

[0031]

[0032] Among them, F i For floor indexing, Z j For a region index, S k For store indexing, H l For shelf indexing, I m For specific product indexes;

[0033] For the customer's location coordinates (x′, y′, z′), the system dynamically calculates them according to the hierarchical indexing rules to achieve a one-to-one correspondence between the customer's location and the location of the target product in the mall.

[0034] As a preferred embodiment of the UWB-based shopping mall guidance system of the present invention, the system dynamically generates optimal guidance routes that avoid high-traffic areas using path planning algorithms and congestion prediction models. It also incorporates a customer behavior prediction model to optimize the routes based on shopping order and dwell habits, and adds weighted constraints of environmental factors during the route generation process.

[0035] The UWB system is used to acquire real-time customer location and trajectory data, while simultaneously collecting current pedestrian traffic in each area. This data is then analyzed based on the customer's historical trajectory and real-time preference feature vector F. t It predicts the set of areas that customers may visit and calculates the expected dwell weight based on dwell time, visit order, and interest probability.

[0036] Introduce environmental weights for each region Taking into account factors such as pedestrian density, aisle width, merchandise display density, and temporary obstacles:

[0037]

[0038] Among them, l j For human flow density, d j The congestion coefficient of the passage, o j Here, a, b, and c represent the obstacle impact index, and a, b, and c are weighting coefficients.

[0039] By combining customer preference weights and environmental weights, each path segment e is defined. ij Comprehensive cost function:

[0040]

[0041] Where, d ij λ1, λ2, and λ3 represent the actual spatial distance of the path segment, and λ1, λ2, and λ3 are adjustment coefficients. Weighting based on expected dwell time;

[0042] Find the optimal path P in the shopping mall spatial graph using a graph search algorithm. * It satisfies the following optimization objectives:

[0043]

[0044] Where P represents the candidate path from the customer's current location to the target set of goods.

[0045] As a preferred embodiment of the UWB-based shopping mall guide system described in this invention, if a loss of connection is detected, a joint compensation algorithm of Kalman filtering and particle filtering is used to dynamically correct the positioning data. Simultaneously, a deep neural network is used to self-learn the positioning error pattern, including:

[0046] Real-time customer location data is acquired from the UWB system, and historical location sequences are recorded. Kalman filtering is used for initial smoothing of the real-time location data. The state prediction formula is as follows:

[0047]

[0048] in, To predict the state, To predict the covariance matrix, A is the state transition matrix, B is the control matrix, and u t-1 For control input, Q is the process noise covariance;

[0049] Then, the prediction results are combined with particle filtering for nonlinear compensation, the particle weights are updated, a deep neural network model is established, the historical positioning error sequence is used as training data, the error pattern mapping is learned, and the joint filtering results are superimposed with the error prediction results of the deep neural network model to obtain the final corrected positioning.

[0050] As a preferred embodiment of the UWB-based shopping mall navigation system described in this invention, the system includes: establishing a navigation credibility index, calculating the confidence level of the compensated location data, and dynamically generating a credibility score, including:

[0051] Obtain customer positioning data after joint filtering and deep neural network error compensation. And record the historical correction data sequence;

[0052] Define the Navigation Credibility Index (CI) t∈[0,1], used to measure the confidence level of the current location data, where 1 indicates complete confidence and 0 indicates location is unavailable; combining location error, filtering residual, and environmental factors, a comprehensive scoring function is defined:

[0053]

[0054] in, σ is the predicted value from the joint filter. t Let η be the historical error variance. t The environmental impact index is represented by A, B, and C, which are adjustment weight coefficients.

[0055] CI is calculated in real time during customer movement. t And adjust the navigation strategy according to the threshold, if CI t <CI th， This triggers path replanning or indicates a low-confidence region, where CI... th This is the preset minimum confidence threshold.

[0056] Secondly, embodiments of the present invention provide a shopping mall guide method based on UWB, including:

[0057] UWB base stations are deployed on each floor and in the passageways of the shopping mall, and centimeter-level precision positioning of customer terminals is achieved through a combination of time difference of arrival and angle of arrival positioning.

[0058] Collect customers' movement speed, dwell time and historical trajectory, and combine them with data from mall WiFi and Bluetooth beacons for multi-source fusion. Use machine learning models to identify customers' shopping preferences and areas of interest in real time.

[0059] Based on the 3D shopping mall building information model and product database, a spatial mapping relationship between product location and UWB coordinate system is established. A hierarchical partition indexing method is adopted to divide the shopping mall into five levels of mapping: floor, area, shop, shelf, and product.

[0060] By using path planning algorithms and congestion prediction models, the optimal shopping guide route that avoids high-traffic areas is dynamically generated. A customer behavior prediction model is introduced to optimize the route based on shopping order and dwell habits, and environmental factors are weighted and constrained during the route generation process.

[0061] When there is electromagnetic interference or signal blockage in the shopping mall, the positioning data is dynamically corrected by using a joint compensation algorithm of Kalman filtering and particle filtering, and the positioning error pattern is self-learned based on deep neural network.

[0062] Establish a navigation credibility index, calculate the confidence level of the compensated positioning data, dynamically generate a credibility score, and automatically trigger alternative route planning or multi-source data weighted fusion when the credibility level is lower than the threshold, so as to realize the switching and protection of the shopping guide route.

[0063] Thirdly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, they implement the steps of the UWB-based shopping mall guide system as described in the first aspect of the present invention.

[0064] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, they implement the steps of the UWB-based shopping mall guide system as described in the first aspect of the present invention.

[0065] The beneficial effects of this invention are as follows: By combining UWB's time difference of arrival (TDOA) and angle of arrival (AOA) joint positioning method, this invention effectively solves the problem of insufficient accuracy in positioning methods such as WiFi and Bluetooth. A five-level spatial mapping index is established to achieve a direct correspondence between customer targets and actual locations, improving the accuracy of product retrieval and navigation in the shopping guide system. Machine learning models are used to model customer dwell time, visit frequency, and promotional interaction behavior, dynamically calculating customer preference distribution, thereby providing personalized recommendations and interest area navigation during the shopping guide process. The introduction of congestion prediction models and customer behavior prediction mechanisms in the path planning process can dynamically avoid high-traffic areas, generate optimal shopping routes, and improve shopping guide efficiency and comfort. Attached Figure Description

[0066] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0067] Figure 1 This is a schematic diagram of the structure of a shopping mall guide system based on UWB.

[0068] Figure 2 This is a flowchart of a shopping mall guide method based on UWB.

[0069] Figure 3 This is a diagram of a computer device for a shopping mall navigation system and method based on UWB. Detailed Implementation

[0070] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0071] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0072] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0073] Example 1

[0074] Reference Figures 1-3 This is the first embodiment of the present invention, which provides a shopping mall guide system based on UWB. The system consists of a positioning module, a data acquisition and fusion module, a model building module, a shopping guide path generation module, an interference compensation construction module, and a credibility assessment module.

[0075] Specifically, the positioning module is used to deploy UWB base stations on each floor and in the passageways of the shopping mall, and to achieve centimeter-level precision positioning of customer terminals through a combination of time difference of arrival and angle of arrival positioning.

[0076] Furthermore, the customer terminal transmits UWB pulses, the base station collects the channel impulse response and performs high-resolution reconstruction of the channel impulse response using sparse reconstruction or multi-resolution time-domain denoising algorithms; the arrival time of the first path is identified using an adaptive threshold or multi-peak peak separation method based on the energy rising edge, and the corresponding time difference arrival observation is calculated, while the original angle of arrival measurement is extracted from the array phase difference or covariance matrix.

[0077] Furthermore, the customer terminal transmits UWB pulse signals with nanosecond-level pulse widths to cover the multipath environment within the shopping mall. Multiple base stations receive the signal and record the channel impulse response sequence.

[0078] A multi-resolution time-domain denoising algorithm is employed, which combines wavelet decomposition with time-domain filtering to process CIR components at different time scales, thereby obtaining a high-resolution CIR representation.

[0079]

[0080] Among them, a i For path gain, τ iLet n(t) be the path delay, and n(t) be the noise residual.

[0081] By dynamically setting a threshold, the rapid rise point of CIR energy is detected, and the first-path delay is extracted. For multi-path near-overlapping scenarios, Gaussian fitting is used to separate adjacent peaks to ensure accurate first-path detection. The arrival times τ of the first path received by multiple base stations are obtained. i Then, the time difference to the observed value Δτ is calculated. ij =τ i -τ j .

[0082] Based on the phase difference information received by the base station antenna array, the incident angle is calculated using the phase difference of the signals received by adjacent antenna elements; the angle of arrival θ is estimated from the covariance matrix of the received signal using the MUSIC or ESPRIT algorithm.

[0083] Furthermore, using the TDOA observation value Δτ ij A hyperbolic constraint is formed to determine the equidistant curves of customer locations; the directional constraint is formed using the AOA observation values ​​θ to obtain the ray direction lines; the two are combined to establish a least-squares optimization problem:

[0084]

[0085] Where p is the customer's location coordinates, b i Let λ be the coordinates of the base station, c be the speed of light, and λ be the weighting factor.

[0086] Specifically, the data collection and fusion module is used to collect customers' movement speed, dwell time, and historical trajectory, and combines it with data from the mall's WiFi and Bluetooth beacons for multi-source fusion. Through machine learning models, it identifies customers' shopping preferences and areas of interest in real time.

[0087] Preferably, a clustering algorithm is used to group customers. The clustering algorithm function is expressed as follows:

[0088]

[0089] Where, μ k For the center of class k, F i Let be the feature vector of the i-th customer;

[0090] A deep time-series prediction model is used to predict the next area of ​​customer interest. The function of the deep time-series prediction model is:

[0091] P(a t+1 |F t =Softmax(W·h) t +b)

[0092] Among them, h tFor the temporally hidden state, P(a t+1 For customers entering area a in the next time slot t+1 The probability distribution;

[0093] Based on real-time input features F t Dynamically calculate the distribution of customer preferences:

[0094]

[0095] Where α, β, γ are weight parameters, Pref j Let Δt represent the real-time probability of a customer's preference for the j-th region. j f is the time the customer spends in the j-th area. j r represents the number of times a customer revisits in the j-th region. j Let Δt be the frequency of customer interaction with promotional information in the j-th area, m be the total number of areas within the mall, and Δt be the frequency of interaction with promotional information in the j-th area. k f is the time a customer spends in the kth region. k r is the number of times a customer revisits in the k-th region. k The frequency of customer interaction with promotional information in the k-th region.

[0096] Furthermore, multiple UWB base stations are deployed within the shopping mall, and customers' portable devices periodically transmit UWB pulse signals. Upon receiving these signals, the base stations process the channel impulse response and, using sparse reconstruction and multi-resolution temporal denoising methods, reconstruct the signal at high resolution, thus obtaining the centimeter-level location coordinates of customers within the mall. Simultaneously, the system collects customer shopping behavior data in the background, including but not limited to: dwell time in various areas; number of times entering the same area; frequency of interaction with promotional information; historical spending amounts and preference records, etc. All data is stored in a structured manner, forming personalized customer characteristic information, providing input for subsequent modeling.

[0097] Furthermore, after acquiring a large amount of customer characteristic data, the system uses clustering algorithms to group customers. By analyzing multi-dimensional features such as dwell time, visit frequency, and promotional interaction frequency, customers are automatically divided into several groups. For example, immersive shoppers spend a long time in specific areas and frequently interact with promotional information; quick-in-quick-out customers spend short time in the mall but have high shopping efficiency and large spending; regular customers repeatedly appear in certain fixed areas, showing stable long-term preferences; and promotion-driven customers have a low overall visit frequency, but show high participation when there are promotional activities. This grouping helps reduce the randomness and noise of the data, enabling subsequent predictive models to be personalized and optimized for different customer groups.

[0098] Furthermore, the system performs time-series modeling of each customer's shopping path, using a time-series prediction model in deep learning to model and predict customer area transitions. This model not only considers the customer's behavior in the current area but also incorporates historical visit trajectories to predict the area the customer is most likely to enter next. For example, if a customer's historical trajectory is "dining area → clothing area → supermarket area," the system predicts that the customer is most likely to enter the "electronics area" next, followed by the "leisure and entertainment area." Compared to traditional methods based solely on statistical frequency, the time-series prediction model can learn more complex behavioral patterns, such as "customers are more likely to enter the clothing area after dining."

[0099] Furthermore, while predicting customer area visit trends, the system dynamically calculates their level of interest in each area based on real-time customer behavior data. This preference distribution comprehensively considers factors such as the length of time customers spend in each area, the frequency of visits, and promotional interactions. For example, a customer might spend 30 minutes in the clothing area, enter the area multiple times, and frequently click on related promotional information; while spending only 15 minutes in the food and beverage area with fewer interactions. After calculating the preference distribution, the system determines that the customer's interest in the clothing area is significantly higher than that in the food and beverage area, and therefore prioritizes pushing clothing-related promotional information when recommending products to sales staff.

[0100] Specifically, the model building module is used to establish a spatial mapping relationship between product location and UWB coordinate system based on the 3D shopping mall building information model and product database. It adopts a hierarchical partition indexing method to divide the shopping mall into five levels of mapping: floor, area, store, shelf, and product.

[0101] Preferably, the points (x, y, z) in the UWB positioning coordinate system are mapped to the BIM spatial coordinate system through a spatial affine transformation matrix M:

[0102] (x′,y′,z′)=M·(x,y,z,1) T

[0103] Where (x′,y′,z′) represents the customer's corresponding location in the BIM space;

[0104] A five-level index structure is used to spatially map customer locations:

[0105]

[0106] Among them, F i For floor indexing, Z j For a region index, S k For store indexing, H l For shelf indexing, I m For specific product indexes;

[0107] For the customer's location coordinates (x′, y′, z′), the system dynamically calculates them according to the hierarchical indexing rules to achieve a one-to-one correspondence between the customer's location and the location of the target product in the mall.

[0108] It should be noted that UWB base stations are deployed in the shopping mall to obtain the three-dimensional coordinates (x, y, z) of customers' UWB tags in real time; through the spatial affine matrix M, these coordinates are mapped to the BIM space to obtain the corresponding position P_BIM of the customer in the BIM coordinate system; the system performs error correction on P_BIM to ensure that the positioning accuracy is within ±0.3 meters.

[0109] Furthermore, floor index F i The system determines the customer's floor based on the z-coordinate (height), for example, the third floor; the area index is Z. j In the BIM model of this floor, it is divided into several areas (such as a catering area, clothing area, digital product area, etc.). The "clothing area" where the customer is located is determined based on the customer's x and y coordinates; store index S k Within the clothing section, retrieve the store boundaries corresponding to the customer's location to determine which women's clothing store the customer has entered; shelf index H l In the store's BIM sub-model, the system determines that the customer is closest to "shelf H3" based on the shelf location closest to their position; Product Index I m The database has a pre-defined relationship between shelves and products. For example, if product "Dress A" is displayed on shelf H3, the system can establish a one-to-one correspondence between customers and target products.

[0110] Furthermore, if a customer selects the target product "Dress A" through the APP, the system can start from the current location P_BIM and call the path planning algorithm to generate the optimal navigation path. During the path, the system combines real-time pedestrian flow data and elevator and escalator usage to dynamically adjust the path to ensure smooth passage. Customers can see their location and path guidance in real time on the APP interface, achieving accurate navigation.

[0111] Specifically, the shopping route generation module uses route planning algorithms and congestion prediction models to dynamically generate the optimal shopping route that avoids high-traffic areas. It also incorporates customer behavior prediction models to optimize the route based on shopping order and dwell habits, and adds weight constraints of environmental factors during the route generation process.

[0112] Preferably, the UWB system is used to acquire real-time customer location and trajectory data, while simultaneously collecting current pedestrian traffic in each area, and then analyzing the customer's historical trajectory and real-time preference feature vector F. t It predicts the set of areas that customers may visit and calculates the expected dwell weight based on dwell time, visit order, and interest probability.

[0113] Introduce environmental weights for each region Taking into account factors such as pedestrian density, aisle width, merchandise display density, and temporary obstacles:

[0114]

[0115] Among them, l j For human flow density, d j The congestion coefficient of the passage, o j Here, a, b, and c represent the obstacle impact index, and a, b, and c are weighting coefficients.

[0116] By combining customer preference weights and environmental weights, each path segment e is defined. ij Comprehensive cost function:

[0117]

[0118] Where, d ij λ1, λ2, and λ3 represent the actual spatial distance of the path segment, and λ1, λ2, and λ3 are adjustment coefficients. Weighting based on expected dwell time;

[0119] Find the optimal path P in the shopping mall spatial graph using a graph search algorithm. * It satisfies the following optimization objectives:

[0120]

[0121] Where P represents the candidate path from the customer's current location to the target set of goods.

[0122] Furthermore, UWB base stations are deployed within the shopping mall. Customer terminals periodically transmit UWB pulse signals. After the base stations collect the signals, they use a joint positioning algorithm to obtain the customer's position (x, y, z) in the UWB coordinate system. The system forms real-time trajectory data of the customer based on continuous position points and stores it in the database. This trajectory data serves as the input for subsequent preference modeling.

[0123] Furthermore, the system retrieves customer historical trajectories and transaction records from the database, extracting features such as dwell time, visit order, and probability of interest, and combines this with the real-time input feature vector F. t The system predicts the set of target areas that customers might visit. For each candidate area, the system calculates the expected dwell weight based on dwell time, visit order, and probability of interest. This weight reflects the attractiveness of different areas to customers.

[0124] Furthermore, the system deploys pedestrian monitoring equipment and sensors in various areas of the shopping mall to collect real-time data on pedestrian density, passageway congestion coefficients, and obstacle impact indices. The system then uses a weighted combination formula to derive the environmental weight for each area. This environmental weight describes the degree to which pedestrian flow, passageways, and obstacles within the current area affect customer passage.

[0125] Furthermore, a spatial topology map is constructed based on the shopping mall's 3D BIM model. Nodes in the map represent areas or key locations, and edges represent passageways or path segments. For each path segment, the system calculates a comprehensive cost function. The system uses an improved graph search algorithm to search for a set P of candidate paths in the topology map and calculates the total cost value of each candidate path based on the comprehensive cost function of the path segment.

[0126] It should be noted that for a path P from the starting point s to the target set (single target t or multiple target sets), the total cost of the path is defined as the sum of the edge weights and node-related costs (depending on the implementation, node dwell costs may be included):

[0127] Cost(P)=sum_{e∈P}C(e)+sum_{v∈VisitNodes(P)}

[0128] NodePenalty(v)

[0129] NodePenalty(v) represents the expected wait and dwell cost to reach node v (included if the customer needs to visit the store to view the products), or the negative of the customer's potential preference gain at that node; when the target is a set of multiple products, Cost(P) includes the dwell cost in the order of access and the cumulative cost required to search for all the targets.

[0130] The system selects the path P with the lowest cost. * This serves as the optimal navigation route for customers. During navigation, if significant changes in area pedestrian density or obstacle index are detected, the system will update the environmental weight E in real time. j This triggers path replanning, thereby generating a new optimal path and ensuring the dynamic adaptability of navigation results.

[0131] Specifically, the interference compensation module is used to dynamically correct the positioning data using a joint compensation algorithm of Kalman filtering and particle filtering when there is electromagnetic interference or signal blockage in the shopping mall. At the same time, it performs self-learning of positioning error patterns based on deep neural networks.

[0132] Preferably, real-time customer location data is obtained from the UWB system, and historical location sequences are recorded. Kalman filtering is used to initially smooth the real-time location data, and the state prediction formula is:

[0133]

[0134] in, To predict the state, To predict the covariance matrix, A is the state transition matrix, B is the control matrix, and u t-1 For control input, Q is the process noise covariance;

[0135] Then, the prediction results are combined with particle filtering for nonlinear compensation, the particle weights are updated, a deep neural network model is established, the historical positioning error sequence is used as training data, the error pattern mapping is learned, and the joint filtering results are superimposed with the error prediction results of the deep neural network model to obtain the final corrected positioning.

[0136] Furthermore, the sampling period (time step) is set to Δt (e.g., 0.1–0.5 s, adjusted according to UWB message frequency and system delay). The measured items obtained or estimated include TOA, AOA, CIR features, SNR, and auxiliary IMU data. Several calibration datasets with ground truth values ​​are collected during the experimental phase. The observed TOA, AOA, CIR sequences, and SNR for this frame are obtained from each base station. NLOS discrimination is performed based on CIR, and the NLOS probability p is calculated. NLOS The measurement covariance matrix R is adaptively set based on SNR and NLOS probabilities. t In NLOS, increase the corresponding entries; convert the observation into a measurement vector z. t .

[0137] Furthermore, in the final correction and confidence control, the PF estimate is superimposed with the DNN error prediction to calculate the positioning confidence index C. I , such as C I Below the threshold CI th Then, one or a combination of the following backoff strategies will be adopted to estimate x using PF. t ; Request multi-source sensor data to participate in fusion; Prompt the user to "Location is unstable"; The result is returned to the upper-level navigation and recommendation modules, and the residual is recorded for subsequent online training of the DNN.

[0138] Specifically, the credibility assessment module is used to establish a navigation credibility index, calculate the confidence level of the compensated location data, dynamically generate a credibility score, and automatically trigger backup route planning or multi-source data weighted fusion when the credibility level is lower than the threshold, so as to realize the switching and protection of the shopping guide route.

[0139] Preferably, customer positioning data is obtained after joint filtering and deep neural network error compensation. And record the historical correction data sequence;

[0140] Define the Navigation Credibility Index (CI) t ∈[0,1], used to measure the confidence level of the current location data, where 1 indicates complete confidence and 0 indicates location is unavailable; combining location error, filtering residual, and environmental factors, a comprehensive scoring function is defined:

[0141]

[0142] in, σ is the predicted value from the joint filter. t Let η be the historical error variance. t The environmental impact index is represented by A, B, and C, which are adjustment weight coefficients.

[0143] CI is calculated in real time during customer movement. t And adjust the navigation strategy according to the threshold, if CI t <CI th， This triggers path replanning or indicates a low-confidence region, where CI... th This is the preset minimum confidence threshold.

[0144] Furthermore, within the sampling period Δt, the system obtains the real-time positioning coordinates X of the customer terminal from the UWB base station. t =(x t ,y t ,z t ), and the predicted position X after joint filtering and DNN compensation. t The system performs a comparison; it calculates the residual value of the customer's current location.

[0145]

[0146] If e t A smaller value indicates a high degree of consistency between positioning and prediction, and strong reliability; if e t If the value is large, then there is a multipath effect or signal blockage;

[0147] The system statistically analyzes the positioning residuals over the most recent N time points and calculates their variance σ. err (t), which reflects the stability of the positioning.

[0148] Furthermore, the system presets a confidence threshold (CI). th (e.g., 0.5), and CI t With CI th Compare; if CI t ≥CI th If the location is reliable enough, continue using the current navigation path; if CI t <CI th This indicates significant uncertainty in positioning, necessitating the triggering of a remedial mechanism; when CI t <CI thWhen entering a low-confidence area, the system invokes the path planning module to regenerate the optimal path based on the currently available high-confidence areas, avoiding low-confidence areas. The system displays or provides a voice prompt on the customer terminal interface: "Current location is unstable, please wait or select a recommended route." Low-confidence areas are marked with a red shading on the map. Upon entering a low-confidence area, the system temporarily adjusts algorithm parameters: increasing the smoothing weight of historical locations to reduce the impact of instantaneous noise; and reducing the residual term e. t The weights used in the credibility calculation make the navigation curve smoother.

[0149] Example 2

[0150] This embodiment is the second embodiment of the present invention, which provides a shopping mall guide system based on UWB. In order to verify the beneficial effects of the present invention, a simulation experiment is conducted for scientific demonstration.

[0151] Specifically, a three-story shopping mall is constructed in a virtual 3D building information model (BIM), with each floor divided into multiple shops, aisles, and shelf areas. Twelve UWB base stations are deployed within the mall, forming a TDOA+AOA hybrid positioning network covering the entire area. A pedestrian flow simulation system is set up to simulate three types of customer flow scenarios: peak hours (lunch / dinner), off-peak hours (morning / afternoon), and low-peak hours (morning / evening).

[0152] Set up 100 virtual customer terminals to simulate three types of shopping behavior patterns: fast shopping: going straight to the target store; leisurely shopping: strolling and stopping in multiple areas; target + impromptu shopping: having a clear goal but stopping due to interest along the way; the system collects customer location data, historical trajectory, traffic flow data and environmental interference data (such as obstacles) in real time.

[0153] Furthermore, to evaluate the performance of the present invention, the following three sets of experiments were designed: Group A (traditional method): Kalman filtering was used only for positioning smoothing; Group B (improved method): joint filtering (Kalman + particle) was used without introducing deep learning and credibility mechanisms; Group C (the method of the present invention): joint filtering + deep neural network error compensation + dynamic adjustment of navigation credibility index + path replanning was used.

[0154] The experimental evaluation metrics include: positioning accuracy, path deviation rate, low confidence area recognition rate, average arrival time, and number of pauses.

[0155] Table 1 Comparison of experimental results using different methods

[0156]

[0157] Table 1 Explanation:

[0158] Group C achieved an average error of 0.22m, enabling centimeter-level positioning, significantly outperforming Groups A and B, thus improving positioning accuracy. Group C's path deviation rate was only 3%, effectively avoiding congested areas, superior to Groups A (12%) and B (7%), demonstrating optimized path planning. Group C achieved a recognition rate of 92%, triggering timely path replanning, significantly better than Group B (65%), and demonstrating accurate identification of low-confidence areas. Group C's average arrival time was reduced by approximately 30%, and the number of pauses decreased by 80%, indicating that the system exhibits higher robustness and intelligence in complex environments.

[0159] Conclusion: Simulation results show that the method of the present invention can significantly improve positioning accuracy, path rationality and navigation reliability in complex shopping mall environments, and has strong practical value and promotion potential.

[0160] This embodiment also provides a computer device suitable for a UWB-based shopping mall guide system, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the UWB-based shopping mall guide system proposed in the above embodiment.

[0161] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0162] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the UWB-based shopping mall guide system proposed in the above embodiments.

[0163] In summary, this invention effectively solves the accuracy problems of WiFi and Bluetooth positioning methods by combining UWB's time difference of arrival (TDOA) and angle of arrival (AOA) joint positioning. A five-level spatial mapping index is established to achieve a direct correspondence between customer targets and actual locations, improving the accuracy of product retrieval and navigation in the shopping guide system. Machine learning models are used to model customer dwell time, visit frequency, and promotional interaction behavior, dynamically calculating customer preference distribution to provide personalized recommendations and interest area navigation during the shopping guide process. The introduction of congestion prediction models and customer behavior prediction mechanisms in the path planning process dynamically avoids high-traffic areas, generating optimal shopping routes and improving shopping guide efficiency and comfort.

[0164] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A UWB-based shopping mall guide system, characterized by, The application relates to a shopping mall customer navigation system, which comprises the following modules: a positioning module, which is used for arranging UWB base stations on each floor and in each aisle of a shopping mall and realizing centimeter-level precision positioning of a customer terminal through a time-difference-of-arrival and angle-of-arrival joint positioning mode; a collection and fusion module, which is used for collecting the moving speed, staying time and historical track of a customer, fusing the data of the shopping mall WiFi and Bluetooth beacons and realizing real-time identification of the shopping preference and attention area of the customer through a machine learning model; a model construction module, which is used for establishing a spatial mapping relationship between the positions of goods and a UWB coordinate system based on a three-dimensional shopping mall building information model and a goods database, adopting a hierarchical partition index mode to divide the shopping mall into five levels of mapping, i.e. floor-region-store-shelf-goods; a shopping guide path generation module, which is used for dynamically generating an optimal shopping guide path that avoids high-flow areas by using a path planning algorithm and a congestion prediction model, introducing a customer behavior prediction model, optimizing the path according to the shopping order and staying habit and superimposing the weight constraints of environmental factors in the path generation process; an interference compensation construction module, which is used for dynamically correcting the positioning data by using a joint compensation algorithm of Kalman filtering and particle filtering when electromagnetic interference or signal blockage exists in the shopping mall and simultaneously self-learning the positioning error mode based on a deep neural network; a credibility evaluation module, which is used for establishing a navigation credibility index, calculating the confidence of the compensated positioning data, dynamically generating a credibility score and automatically triggering a backup path planning or multi-source data weighted fusion to realize switching and guaranteeing of the shopping guide path when the credibility is lower than a threshold value.

2. The UWB-based shopping guide system of claim 1, wherein, The time-difference-of-arrival and angle-of-arrival joint positioning mode realizes centimeter-level precision positioning of the customer terminal, which comprises the following steps: a customer terminal transmits a UWB pulse, a base station collects a channel impulse response and reconstructs the channel impulse response at a high resolution through a sparse reconstruction or a multi-resolution time domain denoising algorithm; an adaptive threshold based on an energy rising edge or a multi-peak separation method is used to identify a first path arrival time, and a time-difference-of-arrival observation value is calculated, and an original angle-of-arrival measurement is extracted from an array phase difference or a covariance matrix.

3. The UWB-based shopping guide system of claim 1, wherein: The machine learning model is used to realize real-time identification of the shopping preference and attention area of the customer, which comprises the following steps: a clustering algorithm is used to group customers, and the clustering algorithm function is represented as follows: where μ k is the center of the kth class, F i is the feature vector of the ith customer; a deep time series prediction model is used to predict the next attention area of the customer, and the deep time series prediction model function is as follows: P(a t+1 |F t ) = Softmax(W · h t +b) where h t is the hidden state at time t, P(a t+1 ) is the probability distribution of the customer entering region a t+1 at the next time slice. According to the real-time input feature F t , the preference distribution of the customer is dynamically calculated: wherein, α, β, γ are weight parameters, Pref j represents the real-time preference probability of the customer to the jth region, Δt j is the stay time of the customer in the jth region, f j is the repeat visit frequency of the customer in the jth region, r j is the promotion information interaction frequency of the customer in the jth region, m is the total number of regions divided in the shopping mall, Δt k is the stay time of the customer in the kth region, f k is the repeat visit frequency of the customer in the kth region, r k is the promotion information interaction frequency of the customer in the kth region.

4. The UWB-based shopping guide system of claim 1, wherein: The three-dimensional shopping mall building information model and the goods database are used to establish a spatial mapping relationship between the positions of goods and a UWB coordinate system, and a hierarchical partition index mode is adopted to divide the shopping mall into five levels of mapping, i.e. floor-region-store-shelf-goods, which comprises the following steps: a point (x, y, z) in a UWB positioning coordinate system is mapped to a BIM space coordinate system through a spatial affine transformation matrix M: (x',y',z') = M - (x,y,z,1) T wherein (x', y', z') is the corresponding position of the customer in the BIM space; a five-level index structure is used to map the customer position in the space: wherein F i is a floor index, Z j is a zone index, S k is a store index, H l is a shelf index, I m is a specific item index; for the positioning coordinates (x', y', z') of the customer, a one-to-one correspondence relationship between the customer position and the target goods position in the shopping mall is realized through dynamic calculation according to the hierarchical index rule.

5. The UWB-based shopping guide system of claim 1, wherein, The path planning algorithm and the congestion prediction model are used to dynamically generate an optimal shopping path that avoids high-traffic areas, a customer behavior prediction model is introduced, the path is optimized according to the shopping order and the habit of staying, and the weight constraint of environmental factors is superimposed in the path generation process, including: Real-time location and trajectory data of customers are acquired by using a UWB system, and current passenger flow of each area is collected. According to the historical trajectory of customers and a real-time preference feature vector F t , a set of areas that the customers are likely to visit is predicted, and expected stay weights are calculated according to stay time, visit order and interest probability; Introducing an environmental weight for each zone Considering the factors of pedestrian density, aisle width, merchandise display density and temporary barriers comprehensively: Wherein, l j is the human flow density, d j is the channel congestion coefficient, o j is the obstacle influence index, a, b, and c are weight coefficients; The customer preference weight and the environmental weight are combined to define a comprehensive cost function for each path segment e ij : wherein d ij is the actual spatial distance of the path segment, λ1, λ2, λ3are adjustment coefficients, is the expected dwell weight; finding an optimal path P in the mall space graph using a graph search algorithm * satisfying the following optimization objectives: Wherein, P represents the candidate path of the customer from the current position to the target commodity set.

6. The UWB-based shopping guide system of claim 1, wherein: If the disconnection is detected, a joint compensation algorithm of Kalman filtering and particle filtering is used to dynamically correct the positioning data, and a deep neural network is used to learn the positioning error pattern based on the deep neural network, including: Real-time positioning data of the customer is obtained from the UWB system, and historical positioning sequences are recorded, the real-time positioning data is preliminarily smoothed by using Kalman filtering, and the state prediction formula is: wherein, is a predicted state, is a predicted covariance matrix, A is a state transition matrix, B is a control matrix, u t-1 is a control input, Q is a process noise covariance; Then, the prediction result is combined with the particle filtering to perform nonlinear compensation, update the particle weight, establish a deep neural network model, use the historical positioning error sequence as the training data to learn the error pattern mapping, superimpose the joint filtering result and the error prediction result of the deep neural network model, and obtain the final corrected positioning.

7. The UWB-based shopping guide system of claim 1, wherein: The navigation credibility index is established to calculate the confidence of the compensated positioning data, and the credibility score is dynamically generated, including: Obtaining customer positioning data after joint filtering and deep neural network error compensation and recording a history of correction data sequences; Definition of navigation confidence index CI t ∈ [0, 1] is used to measure the confidence of the current positioning data, where 1 represents complete confidence and 0 represents that the positioning is unavailable; in combination with the positioning error, the filtering residual and the environmental factors, a comprehensive score function is defined: wherein, is the joint filtered prediction, σ t is the history error variance, η t is the environmental impact index, A, B, C are the adjustment weight coefficients; Real-time computation of CI while the customer is moving t and navigation strategy adjustment according to threshold, if CI t <CI th， trigger path re-planning or hint low confidence area, where CI th is a pre-defined minimum confidence threshold.

8. A UWB-based shopping mall guide method, characterized by, Including: UWB base stations are arranged on each floor and passageway of the shopping mall, and centimeter-level precision positioning of the customer terminal is realized by using the time difference arrival and arrival angle joint positioning method; The moving speed, staying time and historical trajectory of the customer are collected, and the data of the shopping mall WiFi and Bluetooth beacon are combined for multi-source fusion, and the shopping preference and attention area of the customer are identified in real time by using a machine learning model; Based on the three-dimensional shopping mall building information model and the commodity database, a spatial mapping relationship between the commodity position and the UWB coordinate system is established, and a hierarchical partition index method is used to divide the shopping mall into five levels of mapping of floor-region-store-shelf-commodity; The path planning algorithm and the congestion prediction model are used to dynamically generate an optimal shopping path that avoids high-traffic areas, a customer behavior prediction model is introduced, the path is optimized according to the shopping order and the habit of staying, and the weight constraint of environmental factors is superimposed in the path generation process; When there is electromagnetic interference or signal blockage in the shopping mall, a joint compensation algorithm of Kalman filtering and particle filtering is used to dynamically correct the positioning data, and a deep neural network is used to learn the positioning error pattern based on the deep neural network; The navigation credibility index is established to calculate the confidence of the compensated positioning data, and the credibility score is dynamically generated, when the credibility is lower than the threshold, the backup path planning or multi-source data weighted fusion is automatically triggered to realize the switching and protection of the shopping path. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to realize the steps of the UWB-based shopping mall shopping guide system of any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the UWB-based shopping mall shopping guide system of any one of claims 1-7.

Images