Face recognition and microphone-imitating multi-mode collection-based real estate reception behavior intelligent monitoring and analyzing system

Abstract

The application discloses a kind of based on face recognition and imitates microphone multi-mode collection's real estate reception behavior intelligent monitoring and analysis system, belong to real estate case intelligent management field, to solve the problem of fragmentation of prior art, privacy risk is high, hardware deployment rigidification, evaluation dimension single.System distinguishes new and old customers by 90% similarity threshold through entrance face snapshot, and pushes history archives to real estate terminal within 1 second to revisit customer;Adopt hidden collection terminal, integrate noise reduction recording, positioning, bluetooth interaction, reception data is locally encrypted by national encryption algorithm, not uploaded to cloud;Front desk storage cabinet is the only collection node, terminal is returned after decryption upload, after filing, local privacy is automatically erased;Terminal is shared in whole field, single binding;Cloud two sets of AI models respectively output evaluation report to new and old customers, deposit reception technique and customer label.The application gives consideration to privacy protection, hardware reuse and accurate evaluation, and has high integrated practicability.

Description

Technical Field

[0001] This invention relates to the field of intelligent digital management technology for real estate sales sites, and in particular to an integrated full-process monitoring system for sales reception that integrates facial recognition for customer identification at the entrance, a multi-source acquisition terminal with a civilian-style lavalier microphone, UWB high-precision indoor positioning, local offline privacy storage with national cryptographic standards, dedicated data collection for intelligent charging cabinets, dynamic shared equipment scheduling, and differentiated AI reception assessment for different customer groups. Background Technology

[0002] Currently, digital management solutions for sales consultant reception services and customer intention mining in real estate sales offices suffer from industry pain points such as fragmentation, high privacy risks, low equipment utilization, and a single dimension of reception evaluation. Existing published patents can be divided into five categories, but they all only achieve partial and single functions, and there is no complete software and hardware closed-loop solution.

[0003] Existing patents only capture faces using entrance cameras to identify customer visit channels and distinguish between new customers and intermediaries, with the core purpose of channel commission settlement; this has obvious drawbacks:

[0004] (1) The face recognition module operates independently and is not linked with the on-site reception recording, customer tour route, and back-end reception analysis system;

[0005] (2) After identifying repeat customers, only the number of visits is recorded, and no in-depth files such as historical reception dialogues, points of interest, and lingering questions are pushed to the real estate consultant's mobile terminal;

[0006] (3) There is no supporting hardware for voice and spatial data acquisition, making it impossible to conduct differentiated reception quality assessment based on face layering labels.

[0007] Currently available smart employee badges and audio recording terminals for the real estate industry mostly resemble employee badges or name tags in appearance, and their functions are limited to real-time cloud upload of recordings and simple voice keyword extraction. Their shortcomings are as follows:

[0008] (1) The equipment is clearly marked, making it easy for customers to notice the recording equipment, which may cause them to feel resistant and interfere with natural communication; it does not have the concealed appearance design of a civilian microphone;

[0009] (2) The audio data is transmitted to the cloud server in real time throughout the process. Customer conversations and private information are continuously exposed online during the reception process, and there is a lack of local offline encrypted storage architecture;

[0010] (3) Without a UWB indoor positioning module, it is impossible to collect the customer's entire viewing process trajectory and the duration of stay in each functional area. The reception assessment relies solely on a single voice and text, which is one-sided.

[0011] (4) The equipment is fixedly allocated to a single real estate consultant. The equipment is idle during off-peak hours and insufficient during peak hours. The hardware procurement cost is high and the reuse rate is extremely low. There is no shared charging cabinet dynamic allocation and automatic recycling reset mechanism.

[0012] Existing UWB indoor positioning hardware patents mainly focus on UWB + Bluetooth dual-mode hardware applications, such as factory personnel management, museum navigation, and warehouse positioning. However, there are no suitable designs for sales and reception scenarios.

[0013] (1) The noise reduction recording module is not integrated, so it is impossible to collect reception dialogue and spatial behavior data simultaneously;

[0014] (2) There is no supporting intelligent front-end storage and charging equipment, and there is a lack of dedicated business logic for batch local encrypted data reading, collection and transmission;

[0015] (3) It is not connected to the real estate CRM and reception AI analysis platform, so it is impossible to bind the location data with the customer reception file.

[0016] Existing patents for shared charging cabinets and equipment management systems, such as power banks and industrial equipment charging and storage cabinets, only achieve charging and storage location status monitoring; shortcomings:

[0017] (1) It only has power supply function and no privacy protection function such as Bluetooth decryption and reading of local encrypted data packets, batch dedicated line upload, and complete overwrite and erasure of local privacy data;

[0018] (2) There is no dedicated scheduling logic for single temporary binding, reception file mapping, and abnormal equipment warning for sales reception.

[0019] Existing patents for AI analysis software in the sales reception field, which are purely backend algorithm patents, only achieve speech-to-text conversion, customer need tag extraction, and sales script assessment; shortcomings:

[0020] (1) The analysis data source relies solely on a single audio recording text, lacking UWB spatial movement data support, making it impossible to verify the completeness of the reception process;

[0021] (2) It does not distinguish between first-time and repeat customers and uses a uniform evaluation standard, making it impossible to specifically evaluate the quality of follow-up and Q&A for repeat customers;

[0022] (3) The front end has no supporting data acquisition hardware closed loop, and is only an independent software system. When it is deployed, it needs to be equipped with multiple sets of third-party hardware, resulting in poor compatibility.

[0023] In summary, existing technologies are all fragmented single-point functions, lacking a complete integrated system that integrates covert multi-mode data collection, offline local privacy storage, unique node data aggregation, dynamic shared device scheduling, full-link linkage of facial recognition for customer identification, and differentiated AI assessment for different customer groups. The industry urgently needs to solve multiple pain points such as privacy leakage, low device utilization, single dimensions of reception assessment, and lack of pre-requisite support for returning customers. Summary of the Invention

[0024] Purpose of the invention

[0025] To address the shortcomings of existing patented technologies, such as fragmentation, lack of privacy protection, rigid hardware allocation, simplistic reception assessment, and separation of facial recognition and reception processes, this invention provides an intelligent monitoring and analysis system for sales reception behavior based on facial recognition and multi-mode microphone-like data acquisition, achieving the following core solutions:

[0026] It adopts a concealed data acquisition terminal with a civilian-style lavalier microphone to eliminate customers' resistance to monitoring. A single device integrates multi-dimensional data acquisition including recording, UWB positioning, and Bluetooth interaction.

[0027] A national-level local offline storage privacy architecture is constructed, ensuring that the original data throughout the reception process is not uploaded to the cloud. Only the smart charging and storage cabinet serves as the sole compliant node to decrypt and collect the data. Once the upload is complete, all local privacy data on the terminal is automatically erased.

[0028] The terminal is dynamically shared and allocated across the entire venue, without being tied to a fixed real estate consultant, adapting to fluctuations in customer flow, significantly reducing hardware investment and increasing reuse rate;

[0029] The entrance facial recognition system automatically distinguishes between new and returning customers. Returning customers receive a complete historical reception record pushed to their mobile phones in real time, enabling proactive and targeted reception.

[0030] By integrating voice dialogue and UWB guided tour data from multiple sources, the backend AI executes two sets of differentiated quantitative evaluation logics based on the type of customer visit, respectively assessing the completeness of the basic explanation for new customers and the quality of targeted Q&A follow-up for returning customers;

[0031] The five-layer integrated hardware and software closed-loop linkage realizes full-process automation of customer visit identification, dynamic equipment distribution, offline local data collection, encrypted data collection in charging cabinets, and hierarchical AI analysis, filling the gap in existing technology combinations.

[0032] Technical solution

[0033] A smart monitoring and analysis system for sales reception behavior based on face recognition and multi-mode acquisition using a simulated microphone is divided into five layers in a series architecture from top to bottom: face capture and recognition layer, front-end terminal management and control layer, on-site acquisition hardware layer, encrypted and isolated transmission layer, and cloud-based unified AI analysis platform layer;

[0034] The system's complete business loop logic is as follows: When a customer enters the sales office entrance recognition area, a high-definition camera captures their face image without their awareness. Irreversible facial feature values ​​are extracted and compared with a facial database. If the similarity reaches a preset threshold, the customer is considered a returning visitor; otherwise, they are considered a first-time visitor. When a customer is identified as a returning visitor, the cloud-based AI platform immediately sends a pop-up notification to the corresponding sales consultant's mobile app / WeChat account, simultaneously pushing all historical customer records. Front-desk operators complete and input basic information such as customer channel, budget, and preferred apartment type on the unified management platform, binding a unique customer ID to generate an independent reception task. Staff randomly retrieve a fully charged, idle microphone-like acquisition terminal from the front desk's smart charging cabinet, establishing a temporary binding mapping for a single reception via Bluetooth (valid only for this reception, automatically unbinded upon return). The sales consultant wears the terminal and accompanies the customer throughout the viewing and negotiation process, capturing all audio and UWB conversations. The original spatial coordinate data is encrypted and stored in the terminal's local storage unit using national cryptographic algorithms. Throughout the reception process, only lightweight text information such as device battery level and online status is synchronized to the cloud; sensitive original data such as recordings and trajectory data are not transmitted externally. After the reception, the terminal is returned to the storage cabinet. The cabinet hardware verifies the terminal's unique device code. After successful verification, all encrypted data packets are decrypted and read via Bluetooth and transmitted to the cloud platform via a physically isolated encrypted dedicated line. After the data is completely archived, the storage cabinet issues a command to perform a full overwrite and erase of the terminal's local storage, completely erasing all private data from this reception. The terminal automatically enters the fast charging process and returns to an idle and shared state. The cloud platform integrates three types of data: facial recognition visit tags, recorded text, and UWB movement data. Based on the "first visit / return visit" tags, it performs differentiated AI quantitative analysis and outputs corresponding reception evaluation reports, accumulating standardized reception scripts and customer segmentation intention tags.

[0035] The entrance face capture and recognition module deploys high-definition, non-perceptible capture cameras at the main gate of the project site and the visitor entrance of the underground garage, and is equipped with an independent isolated face comparison server. The hardware and office network are physically partitioned, and the supporting functional details are as follows: (1) Face feature extraction and comparison quantification rules: After the camera captures the face, only 128 irreversible encrypted feature values ​​are extracted. The original face capture image is automatically deleted locally 30 seconds after capture and is not permanently stored. The feature library matching similarity threshold is set to 90%: ① Matching similarity < 90%: It is determined to be a first-time visitor. The system automatically generates a new unique ID for the customer and creates a blank file; ② Matching similarity ≥ 90%: It is determined to be a repeat visitor. The system automatically retrieves the customer's full-cycle historical reception data, including the recording text of each visit, the tour route, the value points of concern, price concerns, and unresolved questions.

[0036] (2) Revisit Customer Mobile Terminal Push Interaction Details Identification: Revisit customers trigger push within 1 second. Push carriers support WeChat for enterprises, dedicated property purchase mini-program, and Android / iOS mobile APP. The push content is displayed in three columns: basic visit information (name, number of visits, last visit time), historical attention records (intended unit type, supporting facilities attention points), and a list of follow-up questions. Property purchasers can directly view the complete historical reception assessment report on their mobile phones with one click and prepare targeted explanation plans in advance.

[0037] (3) Face privacy protection details: The face feature database is stored using SM2 national cryptographic asymmetric encryption, and is deployed in an independent database and is not interconnected with the voice and location databases; customers can submit a privacy deletion application with their ID card, and the system can permanently delete all their face feature data with one click in the background, and the system retains the deletion operation log for future reference; face data is only used for visitor recognition and is prohibited from being exported or shared with third parties.

[0038] On-site data acquisition hardware: Imitation civilian lapel clip intelligent multi-mode data acquisition terminal (1) Terminal appearance and structural details The overall shape of the device is completely consistent with the general consumer-grade lapel clip microphone on the market. The size is 28mm×16mm×62mm. The matte plastic shell has no "data acquisition, recording, monitoring" markings. There is only one physical silicone start / stop button on the side. Press it for 0.5s to start recording + UWB positioning data acquisition, and press it for 2s to stop data acquisition. The back has a built-in metal lapel clip that can be clipped to the collar or the edge of the name tag. It is highly concealed and eliminates the customer's vigilance.

[0039] (2) Quantitative parameters of the five major hardware functional modules built into the terminal: ① Dual-microphone high-definition noise reduction recording module: dual silicon microphone array, equipped with adaptive environmental noise reduction algorithm, suppressing 30dB of environmental background noise, sampling rate 48KHz, complete recording of dialogue between the two parties, and accurately identifying the seven major categories of real estate value points keywords: location, school district, commercial, apartment type, property, delivery standard, and preferential policies. ② UWB high-precision positioning module: integrated with DW1000 radio frequency chip, combined with UWB base stations deployed throughout the sales site, indoor positioning accuracy ±10cm, uploading spatial coordinates once per second; the background generates a complete viewing route based on continuous coordinates, and automatically counts the cumulative stay time of customers in the five functional areas of the sand table area, model room, garden landscape, negotiation area, and signing area. ③ Bluetooth 5.3 dual-mode transmission module: supports BLE low power near field interaction, two working modes: temporary binding mode when picking up the device at the front desk, and batch data interaction mode after returning the device; during the reception process, Bluetooth only maintains the heartbeat packet synchronization device status and does not transmit any audio or coordinate raw data. ④ National-level encrypted local storage unit: Equipped with an 8G Flash local storage chip, it uses the SM4 symmetric encryption algorithm to encrypt all recording and location data packets; the device has no external export interfaces such as USB, Type-C, or TF card, making it impossible to copy local data during the reception; only after the smart charging cabinet has passed hardware verification authorization can the encrypted data packets be decrypted and read. ⑤ Long-lasting power supply and control unit: Built-in 320mAh lithium battery, with a continuous data collection time of ≥8 hours on a full charge, meeting the needs of sales reception throughout the day; when the remaining battery is ≤20%, a low battery warning is pushed to the front desk platform. The warning message only includes the device number and battery level, without any original data collected.

[0040] Intelligent charging storage cabinet: This integrated charging storage cabinet, which combines dynamic terminal allocation, recycling charging, and unified data collection, is deployed at the front desk of the sales office. Each cabinet has 27 independent compartments arranged in a neat 3x9 layout. Serving as the sole shared resource pool for all data collection terminals, it integrates fast charging contacts, a Bluetooth communication module, a local data caching unit, and an isolated encrypted transmission network port. Its four core functions are detailed below:

[0041] (1) Dynamic Real-time Sharing and Allocation Mechanism for Equipment: All microphone-like acquisition terminals are universal shared resources throughout the venue and are not bound to any real estate consultant or customer. The system adopts a random dynamic scheduling logic on-site: ① After the front desk completes customer registration and generates a reception task ID, the system automatically filters idle terminals in the cabinet with a battery level ≥30% and no local data uploads. The front desk can randomly take out any terminal; ② After taking out the terminal, Bluetooth automatically establishes a one-time temporary binding relationship of "device hardware code - current reception task ID". The binding lifecycle only lasts for this reception and is automatically released when the customer leaves and the device is returned; ③ After data upload and local privacy data erasure are completed, the terminal is reset to a blank idle resource and can be used by any subsequent visiting customer; It is suitable for peak customer flow, real estate staff rotation, and temporary staffing scenarios. The number of hardware purchases can be reduced by 60%, which greatly reduces investment costs.

[0042] (2) Unique Node Data Aggregation and Multi-layer Privacy Isolation Architecture: The smart charging and storage cabinet is the only hardware node in the entire system with the authority to read, decrypt, and transmit the original collected data. The entire chain is designed with privacy isolation: ① After the terminal is placed in the compartment, the cabinet reads the unique hardware code of the device and verifies it with the whitelist in the background. If the verification fails, reading the local storage is prohibited; if the verification passes, the decryption key is sent via Bluetooth, and all encrypted data packets of the terminal are read to the local cache of the cabinet; ② The cabinet and the cloud AI platform are transmitted through an independent fiber optic encrypted dedicated line. The transmission link is physically isolated from the office WiFi and intranet in the sales office to prevent network eavesdropping; ③ The system sets three levels of data access permissions: only the backend administrator can retrieve the complete original recording and movement data packets; the front desk can only view the device ledger and upload status; the sales consultant can only view the text evaluation report after AI analysis and has no permission to retrieve the original audio and location trajectory. ④ Data Upload and Erasure Rules: After the cloud confirms that the data packet is completely archived, the cabinet issues an overwrite command to perform three rounds of random data overwrite on the local Flash storage of the terminal, completely destroying all original audio and coordinate data of this reception, which cannot be recovered by data recovery tools, and preventing cross-leakage of customer privacy when the equipment is used repeatedly.

[0043] (3) Equipment ledger and abnormal warning function: The storage cabinet synchronizes the number of all terminals in stock, remaining power, usage status, data upload completion record and erase log to the cloud in real time; after the end of business each day, the equipment inventory is automatically performed: if any abnormality occurs, such as terminal not returned, data upload failure, equipment failure or power below 10%, the platform will pop up a red warning window, mark the abnormal equipment number, remind the front desk staff to check, and prevent the loss of reception data and equipment.

[0044] Standardized Execution Steps for the Entire Business Closed Loop

[0045] Step 1 (S1): Entrance facial recognition and instant push notifications for returning customers (recognition time ≤ 1 second, push notification delay ≤ 1 second). When a customer enters the camera recognition area, a facial image is captured, and facial features are extracted and compared to the database; if the similarity is ≥ 90%, a returning visit is determined, and a historical record reminder is immediately pushed to the corresponding property owner's mobile phone; new customers automatically generate a blank unique customer ID.

[0046] Step 2 (S2): Front desk staff complete customer profile creation and binding. Front desk staff enter information such as visit channel, home purchase budget, preferred apartment type, family structure, and number of accompanying persons, and bind the entered form with the customer ID generated by facial recognition to create a unique reception task file.

[0047] Step 3 (S3): On-site dynamic terminal acquisition and one-time temporary Bluetooth binding. Staff randomly select an available terminal, and the temporary mapping between the device and the reception task is completed within 3 seconds via Bluetooth. The terminal simultaneously starts recording and UWB positioning data collection and is ready to be collected. All sensitive raw data throughout the reception process is only encrypted and stored locally, and only lightweight device status messages are synchronized.

[0048] Step 4 (S4): Full-process offline local multi-dimensional data collection. The sales consultant wears a terminal to accompany the client through the entire process of model explanation, showroom tour, and negotiation; the recording module fully records the conversation between the two parties and automatically captures the voice of the seven value points of the property; the UWB module collects spatial coordinates every second and continuously generates the tour route and the duration of stay in each area; all collected materials are encrypted and cached locally on the terminal, without real-time cloud transmission.

[0049] Step 5 (S5): Device recycling and local data packet decryption and collection. After the customer leaves, the terminal is placed in the recycling cabinet slot. The cabinet verifies the terminal's hardware whitelist. After successful verification, Bluetooth decryption reads all encrypted data packets locally on the terminal and caches them in the cabinet.

[0050] Step 6 (S6): Completely overwrite and erase local privacy data on the terminal. After the cloud confirms the data packet reception command, the storage cabinet performs three rounds of random data overwriting on the terminal's local Flash storage, completely destroying the original privacy data of the reception audio and coordinates, automatically unbinding the device from the reception task, and the terminal enters fast charging to return to idle status.

[0051] Step 7 (S7): Encrypted leased line isolated transmission. The encrypted data packets cached in the cabinet are transmitted unidirectionally to the cloud AI platform via a dedicated fiber optic encrypted leased line. The transmission link is physically isolated from the office intranet to prevent data eavesdropping and tampering.

[0052] Step 8 (S8): Cloud platform AI intelligent differentiated quantitative structured analysis. The platform integrates three types of multi-source data: facial visit tags, audio text, and UWB movement lines. It has two sets of independent AI evaluation models built in, and automatically switches the differentiated analysis logic according to the customer's "first visit / return visit" tag. It is equipped with quantitative scoring rules: (1) General basic analysis capabilities: speech to text recognition of property value points, statistics of explanation coverage rate and explanation time of single value points, with a full score of 100 points; based on UWB movement lines to verify the reception process, deducting corresponding scores for missing model rooms, quick tours, and no negotiation communication; capturing customer dialogue keywords, extracting housing needs, price resistance, and unit type preferences, and archiving them into the customer's exclusive file. (2) Differentiated Quantitative Analysis Logic for Different Customer Types ① AI Assessment Model for First-Time Visitors (100 points): The core assessment is the completeness of the explanation of the seven basic value points of the property, with a deduction of 12 points for each missing point; the movement dimension assesses whether the three areas of the model, sample room, and negotiation area are fully shown; and outputs a basic reception standardization rectification list, marking the missing explanation content. ② AI Assessment Model for Repeat Visitors (100 points): The model retrieves all historical reception data of the customer for horizontal comparison, with three core assessment indicators: whether the questions left over from the previous visit are answered in a targeted manner (40 points), whether the detailed policies of the unit types that the customer has been concerned about are supplemented (30 points), and whether preferential solutions are provided for the original price concerns (30 points); the model automatically compares the changes in the movement and focus of the two or more viewings, quantitatively outputs the strength level of the customer's purchase intention (low / medium / high), and outputs optimization suggestions for repeat visit reception. (3) Automatic output of categorized reports: The system generates the "Summary Table of First-Visit Customer Reception Evaluation" and the "Special Evaluation Report of Repeat Customer Reception" daily, and generates monthly tiered reception score statistics for individual real estate consultants. (4) Big data accumulation module for the entire sales site: Automatically distinguishes between high-conversion excellent reception cases of new customers and repeat customers, and accumulates standardized general scripts for first-visit reception, exclusive scripts for follow-up of repeat customers, and standard viewing route templates for different customer groups, which are used for sales team tiered review and pre-job special training. (5) Multi-dimensional customer tiered tag system: Combining the cumulative number of visits by face, multiple rounds of reception dialogue, and spatial behavior data, automatically generates four categories of multi-dimensional customer tags: visit frequency, strength of purchase intention, budget range, and attention to supporting facilities, supporting the long-term refined tiered operation of real estate.

[0053] Beneficial effects

[0054] Compared with the prior art, the present invention has the following significant technical advantages:

[0055] Enhanced concealment reduces customer resistance: Adopting the same appearance design as civilian lavalier microphones, without any monitoring markings, it significantly reduces the sense of surveillance compared to traditional conspicuous name tag-style recording equipment, ensuring a natural reception and communication environment.

[0056] End-to-end privacy protection and enhanced compliance: A complete privacy loop is constructed, consisting of "local encrypted storage on the terminal - unique collection and decryption in the charging cabinet - automatic erasure after uploading". The original data throughout the reception process is not uploaded to the cloud in real time. Compared with existing real-time cloud transmission solutions, this reduces the risk of data leakage at the physical level and meets the requirements for personal information protection compliance.

[0057] Increased hardware reuse rate and reduced investment costs: Terminals are dynamically shared across the entire site and can be temporarily bound once, eliminating the need for one-to-one fixed allocation to real estate consultants. This adapts to fluctuations in customer traffic at the sales site, reducing hardware procurement by more than 60% and solving the problems of insufficient equipment during peak hours and idle resources during off-peak hours.

[0058] Multi-modal data fusion provides a more comprehensive evaluation dimension: Single terminal synchronously collects voice dialogue and UWB spatial movement data. Compared with a single recording analysis solution, it can verify the integrity of the reception process, and the evaluation dimension is expanded from a single text to a dual dimension of "voice + spatial behavior".

[0059] Facial recognition enables seamless customer assessment across the entire process, empowering front-end reception: entrance facial recognition is not only used for customer assessment at the channel level, but also linked to back-end reception records. Follow-up customers receive real-time push of historical data, supporting the preparation of targeted reception plans before property purchase and solving the pain point of the disconnect between customer assessment and reception in the existing system.

[0060] Differentiated AI assessment, tailored to actual business needs: Two independent assessment models are set up for new customers and returning customers, respectively assessing the completeness of basic explanations and the quality of follow-up Q&A. Compared with a uniform standard assessment scheme, this is more in line with the phased business logic of sales reception, and the assessment results are more instructive.

[0061] Integrated closed-loop design with strong compatibility: The five-layer architecture deeply integrates software and hardware, automating the entire process from entry point identification to backend analysis. Compared to fragmented single-point solutions, it eliminates the need for multiple sets of third-party hardware adaptations, resulting in higher deployment efficiency and stronger system stability.

[0062] Data value accumulation supports team growth: Automatically accumulates high-quality reception scripts and standard tour routes for different customer groups, forming reusable sales assets to support tiered team training and capability iteration.

[0063] Eight core innovations of the system

[0064] Innovation of concealed civilian-grade microphone acquisition terminal: The hardware appearance is completely consistent with ordinary consumer-grade lavalier microphones, with no monitoring markings, which greatly reduces customer resistance; existing similar acquisition devices are all conspicuous smart name tags and dedicated recording devices, with a strong sense of monitoring, and lack the concealed appearance design of this invention.

[0065] Offline local national cryptographic encryption + unique node aggregation privacy protection architecture (the biggest differentiating innovation): The entire reception process is recorded, and the original UWB trajectory data is only encrypted and stored locally on the terminal and is not uploaded to the cloud; only the front desk charging and storage cabinet is the only compliant authorized node in the system to decrypt and read the data. After the upload is completed, the local privacy data is completely overwritten and erased in 3 rounds; all existing real estate recording devices adopt a real-time cloud transmission solution, with no local offline storage, authorized aggregation, and automatic erasure of the complete privacy link.

[0066] Terminal dynamic sharing across the entire site, and a single temporary binding scheduling mechanism: The data collection terminal is a public resource across the entire site, randomly allocated on-site, without fixed property binding, and automatically unbound and reset upon return; the existing smart work badges are all fixedly assigned to sales on a one-to-one basis, resulting in high hardware idle rate, high procurement costs, and a lack of shared scheduling logic.

[0067] Single-terminal integrated noise-reducing recording + UWB high-precision positioning + Bluetooth dual-mode interactive multi-mode acquisition: A single small terminal simultaneously collects two types of core data: conversational voice and customer space viewing path; existing UWB positioning hardware lacks a recording module, and the recording badge lacks indoor positioning function, requiring two independent devices to collect data separately, which is cumbersome to deploy.

[0068] Deep integration of facial recognition for customer identification and the entire customer reception process: The entrance camera automatically distinguishes between new and returning customers. If a returning customer is identified, a complete historical reception file is pushed to the buyer's mobile phone within 1 second, providing front-end support for targeted reception. The existing facial recognition customer identification system for sales is only used for channel commission statistics and does not link with the back-end reception data collection and analysis process, thus failing to empower sales reception.

[0069] The intelligent charging cabinet features a multi-functional integrated hardware design: integrating six major functions, including dynamic distribution, automatic fast charging, Bluetooth decryption and reading of encrypted data packets, physical isolation dedicated line transmission, complete erasure of privacy data, and device anomaly warning; existing charging cabinets only have basic charging functions and lack dedicated data processing and privacy destruction logic.

[0070] Dual-model differentiated AI reception quantitative analysis: Two independent scoring and evaluation models are set up for first-time and repeat customers. For repeat customers, historical data is retrieved for comparison to assess the quality of follow-up. Existing reception AI analysis systems only use a uniform evaluation standard, do not distinguish between customer visit types, and cannot measure the targeted nature of repeat service. Five-layer complete hardware and software closed-loop linkage: Face capture and recognition layer → front desk charging cabinet management layer → microphone-like multi-mode acquisition hardware layer → encrypted and isolated transmission layer → layered AI analysis platform, with full-process automated linkage. Existing technologies only implement single-module functions in a fragmented manner, and there is no complete hardware and software integrated business closed loop. Attached Figure Description

[0071] Figure 1 is a five-layer overall architecture diagram of the system; Figure 2 is a schematic diagram of the layout of face capture cameras at the entrance of the sales office; Figure 3 is a schematic diagram of the appearance structure of the civilian-style lapel-type intelligent multi-mode acquisition terminal; Figure 4 is a diagram of the overall structure and storage layout of the intelligent charging and storage cabinet; Figure 5 is a timeline diagram of the entire sales reception process; Figure 6 is a schematic diagram of the overall layout of the UWB base station and the terminal positioning topology at the sales office.

[0072] Figure Label Explanation: 101a, 101b High-definition non-intrusive capture camera; 102 Face comparison isolation server; 103 Customer face recognition sensing area; 104 Sales office entrance gate; 201 Front desk operation terminal; 202 Intelligent charging storage cabinet exterior; 203 Independent charging compartment; 204 Fast charging metal contacts; 205 Miniature Bluetooth antenna; 206 Local encrypted data cache unit; 207 Isolated encrypted transmission network port; 208 Main control board; 209 Equipment ledger / abnormal warning display screen; 301 Imitation civilian lapel clip-type multi-mode acquisition terminal; 302 Microphone; 303 Silicone start / stop physical button; 304 Metal elastic lapel clip; 305 UWB RF positioning unit; 306 Bluetooth 5.3 dual-mode module; 307 SM4 national cryptographic local encrypted storage chip; 308 Lithium battery power supply unit; 401 Physically isolated encrypted dedicated line; 501 Dual-model differentiated AI evaluation module; 502 Big Data Accumulation Module 503 Customer Profile Database 601 UWB Positioning Base Station 602 UWB Wireless Positioning Signal 603 Sand Table Area 604 Showroom Area 605 Garden Landscape Area 606 Negotiation Area 607 Contract Signing Area Detailed Implementation

[0073] The implementation of this system is divided into three phases: hardware deployment, software configuration, and trial operation of business processes. The complete implementation parameters are as follows:

[0074] Hardware deployment phase: Two facial recognition cameras are installed at the entrance of the sales office; one UWB positioning base station is deployed every 80 square meters throughout the entire area; one 27-bay smart charging and storage cabinet is placed at the front desk, along with 20 civilian-style lavalier multi-mode data acquisition terminals; the hardware is networked using independent isolated network cables and physically disconnected from the office network. See Appendix 2 and Appendix 6 for details of the hardware deployment locations.

[0075] Software configuration phase: The face comparison server uses the SM2 / SM4 national cryptographic encryption algorithm, and the face matching similarity threshold is set to 90%; the local storage of the data collection terminal is automatically encrypted, and after the charging cabinet data is uploaded, three rounds of full overwrite and erasure are forcibly performed; two independent reception assessment and scoring models are imported into the cloud AI platform, and a keyword library of seven categories of property value points is configured.

[0076] During the trial operation phase of the business process: Upon arrival, customers are first captured by facial recognition at the entrance. Idle data collection terminals are randomly assigned at the front desk. The entire process of the visit is conducted offline, and the data is automatically collected and erased when the customer leaves. At the end of each business day, two types of reception evaluation reports, namely new customer and repeat visit, are automatically generated in the cloud for the sales team to review and train.

[0077] The entire system operates according to a five-layer architecture, the aforementioned privacy encryption, dynamic scheduling, and layered AI analysis logic, fully realizing the technical effects of covert data collection, privacy protection, efficient hardware reuse, and precise reception and assessment by customer group.

Claims

1. A smart monitoring and analysis system for sales reception behavior based on face recognition and simulated microphone multi-mode acquisition, characterized in that, The architecture consists of five interconnected layers from top to bottom: a face capture and recognition layer, a front-end terminal management and control layer, an on-site data acquisition hardware layer, an encrypted and isolated transmission layer, and a cloud-based unified AI analysis platform layer. The face capture and recognition layer is deployed at the entrance of the sales office and includes a high-definition, non-intrusive capture camera and a face comparison isolation server. The camera only extracts irreversible facial feature values ​​and matches them with a face database. If the similarity reaches a preset threshold, the customer is identified as a repeat visitor; otherwise, the customer is identified as a first-time visitor. When a customer is identified as a repeat visitor, the cloud-based unified AI analysis platform layer pushes a pop-up reminder and the customer's full-cycle historical reception file to the mobile phone of the sales consultant who is following up with the customer. The original face capture image is automatically deleted locally after a limited time after feature extraction, and the facial features are stored independently using an encrypted algorithm. The front-end terminal management layer includes front-end operation terminals and intelligent charging and storage cabinets. The intelligent charging and storage cabinets serve as the only shared resource pool for all data collection terminals. All data collection terminals are not bound to any fixed real estate consultants or customers and are randomly and dynamically allocated on-site. When a terminal is taken out, a temporary binding mapping is established via Bluetooth for a single visit. The binding relationship is only valid for this visit and is automatically released after the device is returned. The storage cabinet is the only hardware node in the entire system with the authority to read, decrypt, and transmit raw data. Terminals can only read local encrypted data packets after being placed in the cabinet and completing hardware whitelist verification. After the cloud confirms the complete archiving of the data, the storage cabinet issues an instruction to perform a full overwrite of the terminal's local storage, completely destroying all original private data from this reception. The on-site data collection hardware layer includes several intelligent multi-mode acquisition terminals with a design resembling civilian lavalier microphones. The terminals integrate a dual-microphone noise-reduction recording module, a UWB high-precision positioning module, a Bluetooth dual-mode transmission module, a local storage unit with encryption algorithms, and a power supply unit. The audio recordings and original UWB spatial coordinate data from the entire reception process are encrypted and stored locally on the terminal. During the reception, only lightweight messages synchronizing device status are sent to the cloud, without transmitting any sensitive original data. The device has no external data export interface, and local encrypted data packets cannot be copied during the reception. The encrypted and isolated transmission layer only opens an independent encrypted dedicated line between the intelligent charging storage cabinet and the unified AI analysis platform in the cloud. The transmission link is physically isolated from the on-site office network. The system sets hierarchical data access permissions, and different roles can only access data within their corresponding permission range. The unified AI analysis platform layer in the cloud has two built-in independent quantitative AI evaluation models that receive facial visit tags, audio recordings, and UWB data collected and uploaded by the charging storage cabinet. The platform utilizes multi-source data flow analysis and automatically switches between differentiated analysis logics: for first-time visitors, the focus is on assessing the completeness of the explanation of the property's basic value points and the standardization of the viewing process across the entire area; for returning visitors, all historical reception data is retrieved for horizontal comparison, and three core indicators are quantitatively assessed: answers to historical questions, supplementary information on unit type policies, and targeted solutions to price concerns, with corresponding quantitative assessment reports generated for each; the platform also includes a big data accumulation module, which differentiates between new and returning customer cases to accumulate standardized reception scripts, and generates multi-dimensional home purchase intention tags for customers based on multi-round reception data.

2. The system according to claim 1, characterized in that, The intelligent multi-mode acquisition terminal, designed to resemble a civilian lavalier microphone, has no recording or monitoring markings on its casing. A metal lavalier clip on the back allows it to be attached to the collar or name tag of a real estate consultant. The side features only one silicone physical start / stop button: a short press simultaneously starts recording and UWB positioning acquisition, while a long press stops acquisition.

3. The system according to claim 1, characterized in that, The Bluetooth dual-mode transmission module has two working modes: a near-field temporary binding mode when picking up the device at the front desk, and a batch encrypted data packet interaction mode after the terminal is returned to the charging cabinet; during the reception process, Bluetooth only maintains low-power heartbeat packets to synchronize the device's online status and power status, and does not transmit audio or coordinate raw data.

4. The system according to claim 1, characterized in that, The face comparison isolation server is independently networked and does not communicate with the voice recording or UWB positioning database; customers can submit a privacy deletion request, and the system can permanently delete all encrypted facial feature values ​​of the user with one click in the background and retain operation logs.

5. The system according to claim 1, characterized in that, The intelligent charging and storage cabinet has a built-in local data cache unit, fast-charging metal contacts, and a compartment Bluetooth communication module. The cabinet synchronizes the number of terminals in stock, remaining power, data upload records, and local data erasure logs to the cloud in real time. When there are terminals not returned, data upload failures, or low battery abnormalities at the end of the business day, the platform will automatically pop up a red warning.

6. The system according to claim 1, characterized in that, UWB positioning base stations are deployed throughout the sales site at a fixed area density. UWB high-precision positioning modules collect terminal spatial coordinates at a fixed frequency. The backend automatically calculates the cumulative time customers spend in the five functional areas: the model area, show flats, garden landscape area, negotiation area, and signing area. This data is used by AI to verify the standardization of the reception process.

7. The system according to claim 1, characterized in that, The cloud-based unified AI analysis platform sets full-score quantitative scoring rules for first-time and repeat customers, automatically generating daily tiered reception summary reports; it distinguishes between high-conversion new customers and repeat customer reception cases to precipitate standardized scripts and standard tour routes for sales team tiered training and review.

8. The system according to claim 1, characterized in that, The preset similarity threshold for facial feature matching is 90%; the push notification delay after identifying repeat customers is ≤1 second; the original facial capture image is automatically deleted locally within 30 seconds after feature extraction.

9. The system according to claim 1, characterized in that, The terminal's local storage uses the SM4 national cryptographic symmetric encryption algorithm, while facial features use the SM2 national cryptographic asymmetric encryption algorithm. After the cloud confirms the data archiving, the storage cabinet performs three rounds of random data overwriting and erasure on the terminal's local storage.

10. The system according to claim 1, characterized in that, The UWB positioning accuracy is ±10cm, and spatial coordinates are collected once per second. The terminal uses a Bluetooth 5.3 dual-mode transmission module, has a built-in 320mAh lithium battery, and can continuously collect data for ≥8 hours on a full charge. The terminal's dimensions are 28mm×16mm×62mm.

11. The system according to claim 1, characterized in that, The smart charging cabinet is equipped with 27 independent charging compartments, arranged in a neat 3 rows and 9 columns. When a user retrieves a device, the system automatically selects idle terminals in the cabinet with a power level of ≥30% and no pending data uploads to allocate the device.

12. The system according to claim 1, characterized in that, The hierarchical data access permissions are set to three levels: only the backend administrator can access the complete original recording and location data package; the frontend can only view the device ledger and data upload status; and the real estate consultant can only view the text evaluation report after AI analysis, and has no permission to access the original audio and location trajectory.

13. A method for intelligent monitoring and analysis of sales reception behavior, based on the system described in any one of claims 1-12, characterized in that, Includes the following steps: S1 Entrance Face Capture and Recognition: Collects facial features of visiting customers, compares them with the feature database to distinguish between first-time and returning customers, and pushes historical reception files to the corresponding sales consultants in real time for returning customers; S2 Customer File Creation and Binding: Enters basic customer information and binds it with a unique facial ID to generate a unique reception task; S3 Dynamic Terminal Retrieval and Temporary Binding: Retrieves idle collection terminals that meet the status requirements from the smart charging and storage cabinet, and establishes a temporary device-task binding relationship for a single reception; S4 Full-Process Offline Collection: The terminal synchronously collects reception dialogue audio and UWB spatial movement data. All raw data is encrypted and stored locally on the terminal, and sensitive raw data is not transmitted to the cloud throughout the process; S5 Device Recycling and Data Collection: After the reception, the terminal is returned to the storage cabinet. After the cabinet completes hardware whitelist verification, it decrypts and reads the local encrypted data packets and caches them; S6 Encrypted Dedicated Line Transmission and Local Data Erasure: The encrypted data packets cached in the cabinet are transmitted to the cloud platform via a physically isolated encrypted dedicated line; After the cloud confirms that the data packet has been completely archived, the storage cabinet issues an instruction to perform a full overwrite and erase of the terminal's local storage, and the terminal is unbound and returns to an idle state; S7 Layered AI Differentiated Analysis: The cloud platform combines customer visit tags and calls the corresponding evaluation model to generate a differentiated reception evaluation report, and accumulates standardized reception data and customer layered tags.

Images