A people flow monitoring and face recognition device based on a vending machine camera

Abstract

This application discloses a vending machine-based crowd monitoring and face recognition device. The device includes a multimodal camera module, a camera control module, a data processing module, and an interactive display module. The multimodal camera module employs a collaborative design of wide-angle, high-definition, and night-vision cameras to achieve panoramic crowd flow acquisition, close-range high-definition face acquisition, and ambient light adaptive illumination. The camera control module dynamically adjusts acquisition parameters based on ambient light intensity and crowd density, achieving time-series synchronization and mode switching. The data processing module integrates the YOLOv8+DeepSORT crowd flow statistics algorithm and the MTCNN+EfficientNet face recognition algorithm, supporting parallel processing and correlation analysis of multi-source data. The interactive display module enables personalized recommendations and operational visualization. This application aims to improve the operational precision and user experience of vending machines through the integration of camera sensing technology and intelligent algorithms.

Description

Technical Field

[0001] This application relates to the fields of smart retail and computer vision technology, and in particular to a people flow monitoring and face recognition device based on a vending machine camera. Background Technology

[0002] Currently, vending machines are widely used in public places, office buildings, communities, and other scenarios, but they have significant technical shortcomings in terms of people flow sensing and user identification:

[0003] Existing vending machines with cameras mostly use a single camera configuration, which needs to simultaneously monitor people flow and recognize faces, making it difficult to balance the field of view and resolution. When monitoring in a panoramic view, the face image is blurry, and when focusing on a face, it cannot cover a large area of ​​people, resulting in insufficient data collection accuracy.

[0004] The camera parameters are mostly fixed, which cannot dynamically adapt to complex lighting environments such as strong light, backlight, and nighttime. It also cannot adjust the acquisition strategy according to the density of people. The acquired images have problems such as overexposure, high noise, and missing targets, which affect the subsequent algorithm processing effect.

[0005] In existing technologies, cameras are used only as independent image acquisition tools and are not deeply integrated with people flow statistics and facial recognition algorithms. The data collection and analysis processes are disconnected, and the hardware performance cannot be fully utilized, resulting in a lack of accurate data support for operational decisions.

[0006] In addition, most devices only implement basic payment authentication or simple people counting functions, and do not utilize the multi-dimensional perception capabilities of cameras to build user profiles and analyze consumption behavior, making it difficult to provide personalized services and resulting in a poor user experience.

[0007] Analysis of existing image recognition devices:

[0008] Single-function camera devices: They can only achieve single functions such as people counting or facial payment, have poor functional scalability, and cannot meet the multi-dimensional data needs of refined operations;

[0009] Fixed parameter camera devices: The acquisition parameters cannot be dynamically adjusted, they have poor environmental adaptability, and the data acquisition accuracy drops significantly in complex scenarios;

[0010] Non-cooperative processing device: The camera and algorithm module work independently, without parameter linkage mechanism, and cannot optimize image acquisition quality through algorithm requirements. Summary of the Invention

[0011] Based on this, this application provides a vending machine camera-based crowd monitoring and face recognition device, which solves the problems of insufficient image acquisition accuracy and poor environmental adaptability in the prior art. It aims to improve the operational precision and user experience of vending machines through the integration of camera sensing technology and intelligent algorithms.

[0012] In a first aspect, a people flow monitoring and facial recognition device based on a vending machine camera is provided. This device is integrated into the vending machine body and includes a multimodal camera module, a camera control module, a data processing module, and an interactive display module, wherein:

[0013] The multimodal camera module includes a wide-angle people flow camera, a high-definition face camera, and a night vision supplementary lighting camera. The three cameras are calibrated to ensure that the acquisition areas overlap and complement each other. The wide-angle people flow camera is installed at the top center of the vending machine to achieve panoramic coverage of a 3-5m range around the vending machine. The high-definition face camera is installed above the front operation panel of the vending machine for clear capture of user faces. The night vision supplementary lighting camera is used for precise supplementary lighting in low light conditions.

[0014] The camera control module connects to the multimodal camera module via the LVDS interface and communicates with the data processing module via the SPI interface. It is used to realize the dynamic adjustment of parameters, mode switching and timing synchronization control of the multimodal camera module, ensuring that the acquisition timing of the three cameras is consistent and that the image data of the same target has the same timestamp.

[0015] The data processing module uses an embedded processor and integrates a people flow statistics module and a face recognition module. It is used to process the image data collected by the multimodal camera module in parallel, and to perform correlation analysis between people flow statistics data and face recognition data to generate user profiles and operation analysis reports.

[0016] The interactive display module is connected to the data processing module via a USB interface and is located on the vending machine's control panel to display personalized recommendations and visual information on pedestrian flow.

[0017] Optionally, in the multimodal camera module:

[0018] The wide-angle crowd camera uses an imaging component with a focal length of 8-12mm, a resolution of ≥4K, and a frame rate of 25-30fps. It is equipped with a dynamic exposure adjustment component and is suitable for lighting environments ranging from 0.1 to 1000 lux.

[0019] The high-definition face camera uses an imaging component with a focal length of 5-8mm, a resolution of ≥2K, and a frame rate of 30fps. It is equipped with a CMOS image sensor with a pixel size of ≥1.4μm and supports automatic focusing on the face area.

[0020] The night vision supplementary light camera is equipped with an 850nm infrared supplementary light component, with a supplementary light distance of 0-5m, 5 levels of supplementary light intensity adjustment, and supports light sensor linkage control. The supplementary light range accurately covers the collection area to avoid supplementary light interference.

[0021] Optionally, the camera control module uses an STM32H743 microcontroller as its core, integrating a light sensor, a people density detection unit, and a parameter adjustment circuit to achieve real-time dynamic adjustment of the acquired parameters.

[0022] The system collects ambient light levels in real time using a light sensor. When the light level is less than 50 lux, the night vision supplementary light camera is automatically turned on and the supplementary light intensity is adjusted according to the light level. When the light level is greater than 500 lux, the camera exposure time is reduced to avoid overexposure.

[0023] The system uses a wide-angle crowd camera to detect crowd density in real time. When the density is greater than 5 people / ㎡, the camera frame rate is increased to 30fps. When the density is less than 1 person / 10㎡, the frame rate is reduced to 15fps to save power.

[0024] When the camera captures abnormal data or the module connection fails, an alarm will be automatically triggered and the fault information will be recorded.

[0025] Optionally, the data processing module supports parallel processing of multi-source data with a response time of ≤100ms, and performs correlation analysis between pedestrian flow statistics and facial recognition data by running a data fusion algorithm to generate user profiles and operation analysis reports.

[0026] Optionally, the crowd flow statistics module operates on the basis of the data processing module, and uses the YOLOv8 target detection algorithm combined with the DeepSORT trajectory tracking algorithm. It uses the panoramic image of the wide-angle crowd flow camera to perform large-scale target detection, and combines the local image of the high-definition face camera to correct the occlusion target recognition deviation. The statistical accuracy is ≥95%, and the detected crowd flow density information is fed back to the camera control module in real time for dynamic adjustment of the acquisition parameters.

[0027] Optionally, the face recognition module operates on the basis of the data processing module, and uses the MTCNN face detection algorithm combined with the EfficientNet feature extraction algorithm to perform distortion correction and contrast enhancement preprocessing on the images captured by the high-definition face camera, and transmits the user identity identifier, biometric features and historical consumption association information output by recognition to the data fusion algorithm.

[0028] Optionally, the camera control module supports single-camera independent operation or three-camera collaborative operation mode, which can be switched via host computer commands; and supports adjusting camera focal length, frame rate, and fill light intensity parameters via remote server or local interactive display module.

[0029] Optionally, the multimodal camera module, camera control module, data processing module, and interactive display module are connected through standardized interfaces, powered by a USB 2.0 standard power supply mode, and internally regulated by 3.3V. Each module can be independently disassembled and replaced to reduce maintenance costs and support functional expansion.

[0030] Optionally, the interactive display module supports the configuration function of adjusting camera parameters through a local interactive panel, and can dynamically adjust the displayed content according to the user profile generated by the data processing module to achieve precise marketing for different user groups.

[0031] In a second aspect, a method for monitoring people flow and recognizing faces based on a vending machine camera is provided, implemented in any of the devices described in the first aspect above, the method comprising:

[0032] Image data is collected through a multimodal camera module: a wide-angle crowd camera collects panoramic images within a 3-5m radius, a high-definition face camera collects face images within a 1-3m close-up area, and a night vision supplementary lighting camera provides supplementary lighting when there is insufficient light. The acquisition sequence of the three cameras is synchronized so that the image data of the same target has the same timestamp.

[0033] The camera control module dynamically adjusts parameters: it collects ambient light intensity in real time and adjusts the supplementary light intensity or exposure time in stages; it detects crowd density in real time and dynamically adjusts the camera frame rate; and it switches between single-camera independent operation and three-camera collaborative operation modes according to the host computer instructions.

[0034] Parallel algorithm processing is performed through the data processing module: the YOLOv8 target detection algorithm combined with the DeepSORT trajectory tracking algorithm is used to detect and track targets in the panoramic images of the wide-angle crowd flow camera; the MTCNN face detection algorithm combined with the EfficientNet feature extraction algorithm is used to perform face recognition on the images of the high-definition face camera; and the crowd flow statistics data and face recognition data are correlated and analyzed to generate user profiles.

[0035] The interactive display module outputs application information: it shows personalized product recommendations based on user profiles and visualizes operational information for hot spots and peak hours based on pedestrian traffic statistics.

[0036] The beneficial effects of the technical solutions provided in this application include at least the following:

[0037] (1) The design of three cameras, namely "wide-angle + high-definition + night vision", is adopted. The wide-angle crowd camera achieves panoramic coverage of 3-5m range, the high-definition face camera focuses on the 1-3m close-range area to ensure clear face capture, and the night vision supplementary light camera adapts to complex lighting environments. The three cameras have overlapping and complementary acquisition areas and synchronized timing, which overcomes the technical defects of existing technologies where a single camera cannot simultaneously meet the requirements of large-scale monitoring and high-definition face capture.

[0038] (2) The camera control module collects ambient light intensity and crowd density in real time, and dynamically adjusts the acquisition parameters such as fill light intensity, exposure time, and frame rate. It can ensure image acquisition clarity in different scenarios such as strong light, backlight, and night. At the same time, it can intelligently switch working modes according to crowd density to balance performance and power consumption, which solves the problems of weak environmental adaptability and unstable image quality of fixed parameter cameras in the existing technology.

[0039] (3) Through bidirectional communication between the camera control module and the data processing module, the image quality requirements and density information are fed back to the camera by the pedestrian flow statistics and face recognition algorithms, which optimizes the image acquisition quality in reverse. At the same time, the pedestrian flow statistics data and face recognition data are correlated and analyzed to generate user profiles, supporting personalized recommendation services. This overcomes the defects of data acquisition and analysis being disconnected and hardware performance not being fully utilized in the existing technology. Attached Figure Description

[0040] To more clearly illustrate the embodiments of this application or the technical solutions in the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0041] Figure 1 A block diagram of a vending machine camera-based crowd monitoring and face recognition device provided in this application embodiment;

[0042] Figure 2 A flowchart illustrating the steps of a method for monitoring people flow and recognizing faces based on a vending machine camera, provided in this application embodiment. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0044] In the description of this application, the terms "comprising," "having," and any variations thereof are intended to cover non-exclusive inclusion, such as a process, method, system, product, or apparatus that includes a series of steps or units, not necessarily limited to those steps or units that are expressly listed, but may also include other steps or units that are not expressly listed but are inherent to these processes, methods, products, or apparatuses, or steps or units added based on further optimizations conceived in this application.

[0045] This application is specifically applied to unmanned operation, user behavior analysis, and personalized service scenarios of vending machines. It aims to improve the operational precision and user experience of vending machines through the integration of camera sensing technology and intelligent algorithms. Specific objectives include:

[0046] It enables precise monitoring of pedestrian flow around vending machines at all times and in all scenarios, and outputs core data such as pedestrian flow, dwell time, and hot spots.

[0047] It enables close-range high-definition face capture and rapid recognition, and outputs user identity information, biometric features, and historical consumption-related information;

[0048] To resolve the conflict between the field of view and resolution of a single camera and improve the accuracy of image acquisition in complex environments;

[0049] To achieve dynamic adaptation of camera parameters to the environment and crowd density, ensuring the stability of data collection in different scenarios;

[0050] Achieve deep collaboration between cameras and algorithm modules to improve the efficiency and accuracy of people flow statistics and facial recognition;

[0051] The modular structure design facilitates the installation, maintenance, and upgrading of the device.

[0052] Please refer to Figure 1 The diagram shows a block diagram of a crowd monitoring and face recognition device based on a vending machine camera provided in this application embodiment. To solve the problems of insufficient image acquisition accuracy and poor environmental adaptability in the prior art, the vending machine body has a reserved installation position to integrate a multimodal camera module, a camera control module, a data processing module and an interactive display module. Each module is connected through a standardized interface, which is convenient for disassembly and maintenance.

[0053] In this embodiment, the multimodal camera module includes a wide-angle crowd camera, a high-definition face camera, and a night vision supplementary lighting camera. The three cameras are calibrated to ensure that their acquisition areas overlap and complement each other. Specifically:

[0054] Wide-angle crowd camera: Installed at the top center of the vending machine, with a field of view ≥120°, achieving panoramic coverage of a surrounding range of 3-5m;

[0055] High-definition face camera: installed at a height of 1.5-1.8m above the control panel, focusing on a close-range area of ​​1-3m to ensure clear capture of the user's face;

[0056] Night vision supplementary lighting camera: Installed in conjunction with the previous two, the supplementary lighting range accurately covers the collection area, avoiding supplementary lighting interference.

[0057] The camera control module connects to the multimodal camera module via the LVDS interface and communicates with the data processing module via the SPI interface. It is used to realize the dynamic adjustment of parameters, mode switching and timing synchronization control of the multimodal camera module, ensuring that the acquisition timing of the three cameras is consistent and that the image data of the same target has the same timestamp.

[0058] The data processing module uses an embedded processor and integrates a people flow statistics module and a face recognition module. It is used to process the image data collected by the multimodal camera module in parallel, and to perform correlation analysis between people flow statistics data and face recognition data to generate user profiles and operation analysis reports.

[0059] The interactive display module is connected to the data processing module via a USB interface and is located on the vending machine's control panel to display personalized recommendations and visual information on pedestrian flow.

[0060] The structure of the core components is shown below:

[0061] (1) Core structure of the multimodal camera module:

[0062] Wide-angle crowd camera: focal length 8-12mm, resolution ≥4K, frame rate 25-30fps, equipped with dynamic exposure adjustment component, adaptable to 0.1-1000lux lighting environment;

[0063] High-definition face camera: focal length 5-8mm, resolution ≥2K, frame rate 30fps, equipped with CMOS image sensor (pixel size ≥1.4μm), supports face area autofocus;

[0064] Night vision supplementary light camera: Equipped with an 850nm infrared supplementary light component, supplementary light distance 0-5m, supplementary light intensity adjustable in 5 levels, and supports light sensor linkage control.

[0065] (2) Camera control module structure:

[0066] Using an STM32H743 microcontroller as the core, it integrates a light sensor, a crowd density detection unit, and a parameter adjustment circuit to achieve real-time dynamic adjustment of the collected parameters.

[0067] (3) Data processing module structure:

[0068] It adopts the NVIDIA Jetson Orin NX embedded processor, integrates people flow statistics algorithm and face recognition algorithm, supports parallel processing of multi-source data, and has a response time of ≤100ms.

[0069] In this embodiment of the application, the core control logic is specifically as follows:

[0070] With the camera control module as the core, dynamic adjustment of parameters and mode switching of multi-modal cameras are realized. The specific control strategy is as follows:

[0071] Environmental adaptation control: The ambient light level is collected in real time by a light sensor. When the light level is <50 lux, the night vision supplementary light camera is automatically turned on and the supplementary light intensity is adjusted according to the light level. When the light level is >500 lux, the camera exposure time is reduced to avoid overexposure.

[0072] Crowd density adaptation control: The crowd density is detected in real time by a wide-angle crowd camera. When the density is >5 people / ㎡, the camera frame rate is increased to 30fps; when the density is <1 person / 10㎡, the frame rate is reduced to 15fps to save power.

[0073] Timing synchronization control: Ensures that the acquisition timing of the three cameras is consistent, and that image data of the same target has the same timestamp, providing a basis for multi-source data fusion.

[0074] The algorithm used in the embodiments is explained in detail below:

[0075] Crowd counting algorithm: The algorithm uses YOLOv8 target detection algorithm combined with DeepSORT trajectory tracking algorithm. It uses panoramic images from a wide-angle camera to perform large-area target detection and combines local images from a high-definition camera to correct the identification deviation of occluded targets. The statistical accuracy is ≥95%.

[0076] Face recognition algorithm: The algorithm uses MTCNN face detection algorithm + EfficientNet feature extraction algorithm to perform distortion correction and contrast enhancement preprocessing on images captured by high-definition cameras. The feature matching response time is ≤0.5 seconds and the matching accuracy is ≥98%.

[0077] Data fusion algorithm: It correlates and analyzes pedestrian flow statistics with facial recognition data to generate user profiles and operational analysis reports, providing data support for personalized recommendations.

[0078] The specific details of the function implementation include:

[0079] Single / Multi-channel Collaborative Control: Supports independent operation of a single camera or collaborative operation of three cameras, which can be switched via host computer commands;

[0080] Configurable parameters: Supports adjusting camera focal length, frame rate, fill light intensity, and other parameters via a remote server or local interactive panel;

[0081] Anomaly detection function: When the camera collects abnormal data or the module connection fails, an alarm will be automatically triggered and the fault information will be recorded.

[0082] In summary, the innovative aspects of this application include:

[0083] Multimodal camera collaborative design: By combining a wide-angle + high-definition + night vision three-camera system, the contradiction between the field of view and resolution of a single camera is resolved, achieving the dual requirements of panoramic monitoring and close-range high-definition acquisition;

[0084] Dynamic parameter adaptation mechanism: Based on real-time detection of ambient light intensity and crowd density, the camera's acquisition parameters and working mode are dynamically adjusted to improve environmental adaptability in complex scenarios;

[0085] Deep hardware and software collaboration: By linking the parameters of the camera control module and the algorithm module, the image acquisition quality is optimized in reverse by the algorithm processing requirements, thereby improving the recognition accuracy.

[0086] Modular structure design: Each functional module is connected with a standardized interface, and can be disassembled and replaced independently, reducing maintenance costs and supporting functional expansion.

[0087] The following are examples of the application effects of this application:

[0088] After deploying this device in a vending machine in the mall's first-floor lobby for two months, the following results were achieved:

[0089] Crowd monitoring: A total of 10,860 people were counted, with an error rate of only 2.8% compared to manual statistics. It successfully identified two peak periods: 8:00-10:00 AM and 3:00-5:00 PM, and accurately located the area in front of the vending machine as a hotspot.

[0090] Face recognition: A total of 862 users have been identified, with a new user registration rate of 65%. Face detection response time is ≤0.3 seconds, matching response time is ≤0.5 seconds, and there have been no recognition errors.

[0091] Environmental Adaptation: The image acquisition clarity meets the algorithm processing requirements in different scenarios such as strong light, backlight, and night, without problems such as overexposure or excessive noise.

[0092] Operational optimization: Based on collected data, operators optimized product display, resulting in a 25% decrease in the out-of-stock rate of best-selling products, a reduction in the average shopping time for users from 60 seconds to 35 seconds, and a 12% increase in repurchase rate.

[0093] Vending machines deployed on the ground floor of the office building:

[0094] By successfully capturing peak passenger flow periods from 8:30-9:30 AM and 12:00-1:00 PM, operators adjusted replenishment frequency accordingly, reducing replenishment costs by 18%.

[0095] By linking user consumption preferences through facial recognition, trendy drinks are recommended to young users, while energy drinks are recommended to business users, resulting in a 15% increase in conversion rate.

[0096] like Figure 2 This application also provides a method for monitoring people flow and recognizing faces based on a vending machine camera, including the following steps:

[0097] Image data is collected through a multimodal camera module: a wide-angle crowd camera collects panoramic images within a 3-5m radius, a high-definition face camera collects face images within a 1-3m close-up area, and a night vision supplementary lighting camera provides supplementary lighting when there is insufficient light. The acquisition sequence of the three cameras is synchronized so that the image data of the same target has the same timestamp.

[0098] The camera control module dynamically adjusts parameters: it collects ambient light intensity in real time and adjusts the supplementary light intensity or exposure time in stages; it detects crowd density in real time and dynamically adjusts the camera frame rate; and it switches between single-camera independent operation and three-camera collaborative operation modes according to the host computer instructions.

[0099] Parallel algorithm processing is performed through the data processing module: the YOLOv8 target detection algorithm combined with the DeepSORT trajectory tracking algorithm is used to detect and track targets in the panoramic images of the wide-angle crowd flow camera; the MTCNN face detection algorithm combined with the EfficientNet feature extraction algorithm is used to perform face recognition on the images of the high-definition face camera; and the crowd flow statistics data and face recognition data are correlated and analyzed to generate user profiles.

[0100] The interactive display module outputs application information: it shows personalized product recommendations based on user profiles and visualizes operational information for hot spots and peak hours based on pedestrian traffic statistics.

[0101] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0102] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A people flow monitoring and face recognition device based on a vending machine camera, characterized in that, The device is integrated into the vending machine body and includes a multimodal camera module, a camera control module, a data processing module, and an interactive display module, wherein: The multimodal camera module includes a wide-angle people flow camera, a high-definition face camera, and a night vision supplementary lighting camera. The three cameras are calibrated to ensure that the acquisition areas overlap and complement each other. The wide-angle people flow camera is installed at the top center of the vending machine to achieve panoramic coverage of a 3-5m range around the vending machine. The high-definition face camera is installed above the front operation panel of the vending machine for clear capture of user faces. The night vision supplementary lighting camera is used for precise supplementary lighting in low light conditions. The camera control module connects to the multimodal camera module via the LVDS interface and communicates with the data processing module via the SPI interface. It is used to realize the dynamic adjustment of parameters, mode switching and timing synchronization control of the multimodal camera module, ensuring that the acquisition timing of the three cameras is consistent and that the image data of the same target has the same timestamp. The data processing module uses an embedded processor and integrates a people flow statistics module and a face recognition module. It is used to process the image data collected by the multimodal camera module in parallel, and to perform correlation analysis between people flow statistics data and face recognition data to generate user profiles and operation analysis reports. The interactive display module is connected to the data processing module via a USB interface and is located on the vending machine's control panel to display personalized recommendations and visual information on pedestrian flow.

2. The apparatus according to claim 1, characterized in that, In the multimodal camera module: The wide-angle crowd camera uses an imaging component with a focal length of 8-12mm, a resolution of ≥4K, and a frame rate of 25-30fps. It is equipped with a dynamic exposure adjustment component and is suitable for lighting environments ranging from 0.1 to 1000 lux. The high-definition face camera uses an imaging component with a focal length of 5-8mm, a resolution of ≥2K, and a frame rate of 30fps. It is equipped with a CMOS image sensor with a pixel size of ≥1.4μm and supports automatic focusing on the face area. The night vision supplementary light camera is equipped with an 850nm infrared supplementary light component, with a supplementary light distance of 0-5m, 5 levels of supplementary light intensity adjustment, and supports light sensor linkage control. The supplementary light range accurately covers the collection area to avoid supplementary light interference.

3. The apparatus according to claim 1, characterized in that, The camera control module uses an STM32H743 microcontroller as its core, integrating a light sensor, a people density detection unit, and a parameter adjustment circuit to achieve real-time dynamic adjustment of the collected parameters. The system collects ambient light levels in real time using a light sensor. When the light level is less than 50 lux, the night vision supplementary light camera is automatically turned on and the supplementary light intensity is adjusted according to the light level. When the light level is greater than 500 lux, the camera exposure time is reduced to avoid overexposure. The system uses a wide-angle crowd camera to detect crowd density in real time. When the density is greater than 5 people / ㎡, the camera frame rate is increased to 30fps. When the density is less than 1 person / 10㎡, the frame rate is reduced to 15fps to save power. When the camera captures abnormal data or the module connection fails, an alarm will be automatically triggered and the fault information will be recorded.

4. The apparatus according to claim 1, characterized in that, The data processing module supports parallel processing of multi-source data with a response time of ≤100ms. It also uses a data fusion algorithm to correlate and analyze pedestrian flow statistics and facial recognition data to generate user profiles and operational analysis reports.

5. The apparatus according to claim 1, characterized in that, The crowd flow statistics module operates on the basis of the data processing module. It uses the YOLOv8 target detection algorithm combined with the DeepSORT trajectory tracking algorithm. It uses the panoramic image of the wide-angle crowd flow camera to perform large-scale target detection, and combines the local image of the high-definition face camera to correct the recognition deviation of occluded targets. The statistical accuracy is ≥95%, and the detected crowd flow density information is fed back to the camera control module in real time for dynamic adjustment of the acquisition parameters.

6. The apparatus according to claim 1, characterized in that, The face recognition module operates on the basis of the data processing module. It uses the MTCNN face detection algorithm combined with the EfficientNet feature extraction algorithm to perform distortion correction and contrast enhancement preprocessing on the images captured by the high-definition face camera, and transmits the user identity, biometric features and historical consumption association information output by recognition to the data fusion algorithm.

7. The apparatus according to claim 1, characterized in that, The camera control module supports single-camera independent operation or three-camera collaborative operation mode, which can be switched via host computer commands; and supports adjusting camera focal length, frame rate, and fill light intensity parameters via remote server or local interactive display module.

8. The apparatus according to claim 1, characterized in that, The multimodal camera module, camera control module, data processing module, and interactive display module are connected through standardized interfaces. The power supply adopts the USB 2.0 standard power supply mode and is internally regulated with 3.3V. Each module can be independently disassembled and replaced to reduce maintenance costs and support functional expansion.

9. The apparatus according to claim 1, characterized in that, The interactive display module supports the configuration function of adjusting camera parameters through a local interactive panel, and can dynamically adjust the displayed content according to the user profile generated by the data processing module, so as to achieve precise marketing for different user groups.

10. A method for monitoring people flow and recognizing faces based on a vending machine camera, implemented in the device described in any one of claims 1-9, characterized in that, The method includes the following steps: Image data is collected through a multimodal camera module: a wide-angle crowd camera collects panoramic images within a 3-5m radius, a high-definition face camera collects face images within a 1-3m close-up area, and a night vision supplementary lighting camera provides supplementary lighting when there is insufficient light. The acquisition sequence of the three cameras is synchronized so that the image data of the same target has the same timestamp. The camera control module dynamically adjusts parameters: it collects ambient light intensity in real time and adjusts the supplementary light intensity or exposure time in stages; it detects crowd density in real time and dynamically adjusts the camera frame rate; and it switches between single-camera independent operation and three-camera collaborative operation modes according to the host computer instructions. Parallel algorithm processing is performed through the data processing module: the YOLOv8 target detection algorithm combined with the DeepSORT trajectory tracking algorithm is used to detect and track targets in the panoramic images of the wide-angle crowd flow camera; the MTCNN face detection algorithm combined with the EfficientNet feature extraction algorithm is used to perform face recognition on the images of the high-definition face camera; and the crowd flow statistics data and face recognition data are correlated and analyzed to generate user profiles. The interactive display module outputs application information: it shows personalized product recommendations based on user profiles and visualizes operational information for hot spots and peak hours based on pedestrian traffic statistics.

Images