An access control permission intelligent control method and system based on an internet of things

Abstract

The application relates to the technical field of biometric recognition, and discloses a door access permission intelligent control method and system based on the Internet of Things, which mainly solves the technical problem that the face and fingerprint living body detection of the existing door access system are independent of each other and are easily attacked and cheated by physical combination such as a 3D mask combined with a false fingerprint film. Mainly comprising: based on a global clock synchronization mechanism, collecting a face video stream and a fingertip optical reflection sequence containing a unified time identifier; respectively extracting photoelectric response data of a face region and a fingertip region to generate a face rPPG signal and a fingertip PPG signal; based on the unified time identifier, performing time sequence synchronization on the two groups of pulse wave signals to determine the physiological conduction time sequence relationship and waveform consistency between the face and the fingertip; when a preset physiological living body authentication condition is met, generating a living body authentication identifier; finally, in response to the identifier, jointly authenticating an identity feature and outputting a door access instruction.

Description

Technical Field

[0001] This invention relates to the field of biometric identification technology and discloses an intelligent access control method and system based on the Internet of Things. Background Technology

[0002] With the development of IoT technology, multimodal access control systems based on facial and fingerprint recognition have been widely used. However, their liveness detection mechanisms are often isolated, operating in a "modal silo" state. In existing identity verification processes, facial liveness detection and fingerprint liveness detection are handled by separate algorithm modules, with only a logical overlap between them. This fragmented anti-spoofing architecture poses a significant security vulnerability. Attackers can easily perform combined deception attacks by physically piecing together components. For example, by simultaneously wearing a high-precision 3D silicone face mask and using a highly realistic fake fingerprint film engraved with legitimate textures, they can bypass the independent liveness detection modules, thus easily breaching high-security access control defenses.

[0003] The root cause of the aforementioned security vulnerabilities lies in the fact that traditional liveness detection schemes only rely on superficial comparisons of physical textures, surface temperatures, or specific localized movements, completely severing the inherent physiological linkages inherent in the real human body as a holistic living system. In summary, existing access control authentication technologies lack in-depth analysis of the physiological transmission timing and pulse waveform consistency between different biometric features such as the human face and fingertips, failing to achieve cross-modal, low-level physiological rhythm cross-verification. Consequently, they cannot effectively resist high-dimensional physical combination forgery attacks, necessitating the introduction of a novel liveness detection mechanism to address this security vulnerability. Summary of the Invention

[0004] To overcome the aforementioned shortcomings of the prior art and to achieve the above objectives, the present invention provides the following technical solution: an intelligent access control method based on the Internet of Things, comprising: S1. Obtain the authentication trigger signal and collect a facial video stream and fingertip optical reflection sequence containing a unified time identifier based on the global clock synchronization mechanism; S2. Extract physiological signals from the facial region in the facial video stream to generate facial photoplethysmography (PPG) signals, and perform optical response analysis on the fingertip optical reflection sequence to generate fingertip PPG signals. S3. Based on the unified time identifier, the facial photoplethysmography (PPG) wave signal and the fingertip PPG wave signal are time-synchronized, the corresponding pulse change features are extracted, and the physiological conduction timing relationship and waveform consistency between the two are determined. S4. Determine whether the preset liveness authentication conditions are met based on the physiological conduction timing relationship and waveform consistency. If the conditions are met, generate a liveness authentication identifier. S5. In response to the liveness authentication identifier, perform joint authentication on the identity features in the facial video stream and the fingertip optical reflection sequence, and generate an access control command when the authentication is successful.

[0005] Preferably, step S1, which involves acquiring a facial video stream and a fingertip optical reflection sequence containing a unified time identifier based on a global clock synchronization mechanism, specifically includes: Receive the authentication trigger signal sent by the access control terminal and extract the basic time information corresponding to the authentication trigger signal; Based on the basic time information and the device response delay information corresponding to the facial image acquisition device and the fingertip optical acquisition device, the time reference of different acquisition devices is calibrated to generate a unified time identifier. Based on the unified time identifier, the facial image acquisition device is controlled to acquire continuous facial video frames, and the fingertip optical acquisition device is simultaneously controlled to acquire optical reflection data within the corresponding time period.

[0006] Preferably, step S2 involves extracting physiological signals from the facial region in the facial video stream, specifically including: Face detection and feature localization are performed on consecutive image frames in the facial video stream to obtain positional change information of key facial regions; Based on the key facial regions, an effective area for physiological signal acquisition is determined, including the forehead region and the cheek region; Color change information corresponding to multiple consecutive image frames within the effective area is extracted, and the color change information is converted into a time series to generate a facial photoplethysmography (PPG) signal reflecting changes in facial blood volume.

[0007] Preferably, the step of generating the facial photoplethysmography (PPG) signal in S2 includes the steps of generating an initial signal and suppressing interference, specifically including: Acquire interference information caused by changes in ambient light, facial posture, and facial expression during the facial video stream acquisition process; The initial signal is corrected based on the interference information to reduce the influence of non-physiological factors on the signal change trend, thereby obtaining the facial photoplethysmography signal.

[0008] Preferably, the optical response analysis of the fingertip optical reflection sequence in step S2 specifically includes: The fingertip optical acquisition device is controlled to emit detection light of a preset wavelength towards the fingertip area, and the reflection response data of the fingertip area to the detection light is acquired. Based on the light intensity change information corresponding to different sampling times, the characteristics of blood volume change in the fingertip are extracted, and the optical response data obtained by continuous acquisition are processed in a time sequence to generate the fingertip photoplethysmography (FP-P) signal.

[0009] Preferably, determining the physiological conduction timing relationship in S3 specifically includes: Periodic features in the corresponding pulse change features are extracted from the facial photoplethysmography (PPG) signal and the fingertip PPG signal, respectively, and a time correspondence between the two pulse signals is established based on the unified time identifier. Based on the order of appearance of the periodic change characteristics in the two pulse wave signals, the pulse conduction direction between the facial region and the fingertip region is determined, and it is judged whether the pulse conduction direction conforms to the physiological time sequence relationship corresponding to the human blood circulation process.

[0010] Preferably, determining the waveform consistency in S3 specifically includes: According to a preset time window, periodic change segments in the facial photoplethysmography (PPG) signal and the fingertip PPG signal are obtained respectively. The correlation analysis is performed on the changing trends, periodic characteristics and fluctuation patterns of the two pulse wave signals within the corresponding time window. The consistency of the waveforms is determined based on the degree of synchronous change between the two signals, so as to determine whether the two pulse wave signals originate from the same human physiological activity.

[0011] Preferably, before generating the liveness authentication identifier in step S4, an abnormal authentication status judgment step is included, specifically including: Real-time detection of the temporal correspondence and waveform changes between the facial photoplethysmography (PPG) signal and the fingertip PPG signal; When there is no time relationship between two pulse wave signals that conforms to the physiological conduction law of the human body, or when the changing trend of the two pulse wave signals does not meet the preset consistency condition, it is determined that there is an abnormal state in the current authentication process, and the generation of the liveness authentication identifier is prevented.

[0012] Preferably, the identity feature joint authentication in S5 specifically includes: Facial structural feature information is extracted from the facial video stream that has passed liveness authentication, and fingerprint texture feature information is extracted from the fingertip image data corresponding to the fingertip optical reflection sequence; The facial structure feature information and the fingerprint texture feature information are respectively matched with the authorized identity information pre-stored in the access control system; When both the facial structure feature information and the fingerprint texture feature information meet the corresponding authentication conditions, the access control command is generated.

[0013] An IoT-based intelligent access control system includes: The synchronous acquisition module acquires the authentication trigger signal and acquires a facial video stream and fingertip optical reflection sequence containing a unified time identifier based on a global clock synchronization mechanism; The signal extraction module extracts physiological signals from the facial region in the facial video stream to generate facial photoplethysmography (PPG) signals, and performs optical response analysis on the fingertip optical reflection sequence to generate fingertip PPG signals. The timing alignment module performs timing synchronization of the facial photoplethysmography (PPG) wave signal and the fingertip PPG wave signal based on the unified time identifier, extracts the corresponding pulse change features, and determines the physiological conduction timing relationship and waveform consistency between the two. The liveness determination module determines whether the preset liveness authentication conditions are met based on the physiological conduction timing relationship and waveform consistency. When the conditions are met, a liveness authentication identifier is generated. The access control module, in response to the liveness authentication identifier, performs joint authentication of the identity features in the facial video stream and the fingertip optical reflection sequence, and generates an access control command when the authentication is successful.

[0014] Compared with the prior art, the present invention has the following beneficial effects: This invention establishes a linkage mechanism between facial and fingertip pulse wave signals at the underlying level, making the liveness detection of these two biometric features no longer a separate, independent process. This holistic cross-verification mechanism prevents attackers from using physical methods such as combining 3D face masks with fake fingerprint films to deceive different sensors separately, fundamentally resisting high-dimensional combined forgery attacks and significantly improving the anti-counterfeiting ceiling of access control systems.

[0015] This invention abandons the superficial comparison method of traditional solutions that rely solely on surface texture or local temperature, and instead delves into the temporal continuity of the blood circulation system within the same human body. By accurately comparing the temporal relationship and waveform consistency of pulse conduction between the face and fingertips, it utilizes the unforgeable internal physiological conduction laws of living organisms, making the basis for liveness determination more profound and reliable, and solving the defect of traditional liveness detection being easily penetrated by high-fidelity molds.

[0016] This invention forcibly binds facial video streams to fingertip optical sequences using a unified time identifier from the very beginning of data acquisition, ensuring absolute consistency of multimodal data on a temporal reference basis. This architecture not only provides a solid data foundation for the accurate calculation of subsequent physiological time-series relationships and avoids misjudgments or omissions caused by response delays from different acquisition devices, but also significantly enhances the stability and fault tolerance of the IoT access control system. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the process of an IoT-based intelligent access control system according to the present invention. Figure 2 This is a schematic diagram of an IoT-based intelligent access control system according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example

[0019] Please see Figure 1 As shown, this embodiment provides an intelligent access control method and system based on the Internet of Things, specifically including the following steps: S1. Obtain the authentication trigger signal and collect a facial video stream and fingertip optical reflection sequence containing a unified time identifier based on the global clock synchronization mechanism; S2. Extract physiological signals from the facial region in the facial video stream to generate facial photoplethysmography (PPG) signals, and perform optical response analysis on the fingertip optical reflection sequence to generate fingertip PPG signals. S3. Based on the unified time identifier, the facial photoplethysmography (PPG) wave signal and the fingertip PPG wave signal are time-synchronized, the corresponding pulse change features are extracted, and the physiological conduction timing relationship and waveform consistency between the two are determined. S4. Determine whether the preset liveness authentication conditions are met based on the physiological conduction timing relationship and waveform consistency. If the conditions are met, generate a liveness authentication identifier. S5. In response to the liveness authentication identifier, perform joint authentication on the identity features in the facial video stream and the fingertip optical reflection sequence, and generate an access control command when the authentication is successful.

[0020] The acquisition of facial video streams and fingertip optical reflection sequences, which contain a unified time stamp, is based on a global clock synchronization mechanism. Specifically, this includes: The system receives the authentication trigger signal from the access control terminal and extracts the corresponding basic time information. In practical applications, when a user's limb approaches the access control terminal and triggers the microwave radar, or when a user touches the access control button and triggers the capacitive sensor, the hardware interrupt pin of the access control main control chip will capture a rising edge signal, which is then identified by the system as the authentication trigger signal. Upon receiving this signal, the main control chip reads the local real-time clock system kernel timestamp and uses this absolute time node as the basic time information.

[0021] Based on fundamental time information and the device response latency information of facial image acquisition devices and fingertip optical acquisition devices, the time base of different acquisition devices is calibrated to generate a unified time identifier. Due to the different hardware characteristics of facial image acquisition devices (such as RGB cameras) and fingertip optical acquisition devices (such as under-display optical fingerprint modules), there are different response latencyes between receiving a trigger command and outputting the first frame of valid data. The response latency of facial image acquisition devices typically includes sensor power-on time and first frame exposure time; while the response latency of fingertip optical acquisition devices typically includes LED light source illumination time and photodiode integration time.

[0022] To achieve strict synchronization, the main control chip first compares the response latency information of the two devices and selects the device with the longest latency. To prevent microsecond-level instruction blocking in the underlying bus communication, the system also adds a preset system clock jitter margin coefficient to the longest latency. Specifically, the system calculates the unified time identifier using the following timing calibration formula: ,in, This indicates the unified time identifier generated after calibration, which is the future absolute time point that triggers the data collection uniformly; This indicates the basic time information obtained; This indicates the response delay time of the facial image acquisition device; Indicates the response delay time of the fingertip optical acquisition device; This is a function to find the maximum value. This is the preset system clock jitter margin factor. The main control chip will calculate the... This absolute point in time in the future serves as the calibrated unified time marker.

[0023] For example, suppose the access control terminal captures the user's trigger signal at an absolute time of 08:00:00.000 (accurate to milliseconds). This time is the base time information. The system's factory calibration indicates that the response latency of the facial image acquisition device is 30 milliseconds, and the response latency of the fingertip optical acquisition device is 15 milliseconds. Simultaneously, the system's preset jitter margin is 5 milliseconds. During calibration, the system compares and determines that the longest latency is 30 milliseconds for the facial device. The system then adds the base time information (08:00:00.000), the longest latency of 30 milliseconds, and the margin factor of 5 milliseconds. The resulting unified time identifier is set to the 35th millisecond after triggering, i.e., the absolute time 08:00:00.035.

[0024] The unified time stamp controls the facial image acquisition device to acquire continuous facial video frames, and simultaneously controls the fingertip optical acquisition device to acquire optical reflection data within the corresponding time period.

[0025] The main control chip sends synchronization control commands, including the aforementioned unified time identifier, to the underlying drivers of the facial image acquisition device and the fingertip optical acquisition device. Upon receiving the commands, both devices enter a standby ready state and use internal high-frequency timers to count down. When the system time precisely reaches the physical moment corresponding to the unified time identifier, both acquisition devices trigger their first data sampling simultaneously. During subsequent continuous acquisition, the main control chip uses this unified time identifier as the header of each data frame, binding it to each frame of image and optical data. This ensures, at the underlying physical level, that the facial video stream and the fingertip optical reflection sequence have an absolutely consistent time reference, eliminating hardware asynchronous interference for subsequent extraction of time-synchronized physiological pulse signals.

[0026] Physiological signals are extracted from the facial region in the facial video stream, specifically including: Face detection and feature localization are performed on consecutive image frames in a facial video stream to obtain positional change information of key facial regions. After acquiring the facial video stream, the main control chip sequentially feeds consecutive image frames into a pre-deployed facial landmark detection model, such as a high-precision face alignment network containing multiple feature points. The model identifies key locations such as the facial contours, corners of the eyes, tip of the nose, and corners of the mouth, and outputs the specific coordinate values ​​of these locations in the two-dimensional pixel coordinate system of each frame. Over time, the dynamic displacement trajectory of these coordinate values ​​constitutes the positional change information of key facial regions. The system can then track the spatial pose of the face in real time, providing a coordinate reference for subsequent resistance to motion artifacts caused by slight head movements.

[0027] The system determines the effective area for physiological signal acquisition based on key facial regions, including the forehead and cheeks. Because capillary density varies across different facial areas, and facial expressions such as blinking and mouth opening and closing significantly interfere with the extraction of optical volume changes, the system uses geometric calculations based on the extracted key point coordinates to automatically select the relatively flat, smaller, and more subcutaneously vascularized central forehead region, as well as the left and right cheek regions, avoiding the jaw muscles. These three sub-regions are marked as regions of interest and together constitute the effective area for physiological signal acquisition.

[0028] The system extracts color change information from multiple consecutive image frames within the effective region and performs time-series transformation on this information to generate a facial photoplethysmography (PPG) signal reflecting changes in facial blood volume. Because hemoglobin in the blood has the strongest absorption characteristic for the green band of visible light, the blood volume in the subcutaneous blood vessels of the face changes regularly with the periodic pumping of the heart, resulting in slight periodic fluctuations in the intensity of reflected light from the face. The system extracts all pixels within the effective region of each frame and separates the color channel most sensitive to blood changes. The spatial average of the intensity values ​​of all pixels in this channel is calculated as the representative color change information for that frame. The system concatenates the representative color change information from each frame arranged in chronological order in the video stream, thus converting the discrete image frame data into a one-dimensional numerical vector that fluctuates continuously over time; this vector is the facial PPG signal.

[0029] For example, suppose a door access camera captures a 2-second facial video stream at a frame rate of 30 frames per second, totaling 60 frames. In the first frame, the model detects a face and, using the bridge of the nose as an anchor point, translates upwards by a specific pixel distance to select a 40×20 pixel forehead region, and translates to the sides to select two 30×30 pixel cheek regions. These three regions contain a total of 2600 pixels.

[0030] The system extracts the color values ​​of these 2600 pixels in the green channel, sums them, and calculates the average color value, assuming the calculated average color value is 145.5. As the video stream progresses, the system tracks and calculates the average green channel color value of the remaining 59 frames in the same area. Due to changes in blood volume caused by the heart pumping blood, these average color values ​​will fluctuate around a baseline value.

[0031] The system concatenates these 60 average color values ​​generated in chronological order to form a one-dimensional time series containing 60 values. This data sequence, which records the subtle color fluctuations, constitutes a 2-second facial rPPG signal that can be used for subsequent liveness detection.

[0032] Generating facial photoplethysmography (PPG) signals includes steps such as generating an initial signal and performing interference suppression, specifically including: The system acquires interference information caused by changes in ambient light, facial posture, and facial expression during the acquisition of facial video streams. During the initial signal acquisition process, the system simultaneously monitors three typical non-physiological interference sources. The first is ambient light variation interference, such as flickering corridor lights or cloud cover blocking sunlight. This interference causes the brightness values ​​of the red, green, and blue color channels of the facial image to fluctuate drastically in the same proportion at the same moment. The second is facial posture variation interference, where slight head movements cause spatial displacement of the extraction area, manifesting as baseline drift on the signal time axis. The third is facial expression variation interference, such as blinking or smiling, which causes momentary contractions of facial muscles, appearing as short-duration high-frequency spikes in the signal waveform. The system identifies these abnormal fluctuations as interference information by comparing the amplitude abrupt change rate of the initial signal waveform and the interconnected fluctuation characteristics of the multi-color channels.

[0033] The initial signal is corrected based on interference information to reduce the influence of non-physiological factors on signal variation trends, resulting in a facial photoplethysmography (PPG) signal. For identified interference, the system employs a multi-level filtering and separation strategy for correction. Regarding ambient light variations, the system uses a blind source separation algorithm, assuming that the actual physiological pulse signal and the external ambient light signal are two independent signal sources. By extracting data features from multiple color channels, the mixed initial signal is decoupled, and interference components strongly correlated with ambient light fluctuations are extracted, retaining only the physiological feature components primarily driven by changes in microvascular blood volume.

[0034] To address baseline drift and high-frequency spikes caused by changes in facial posture and expression, the system employs a bandpass filter in the frequency domain for truncation. Since normal human heart rate typically falls within a fixed range, the system strictly sets the passband of the bandpass filter within the frequency range corresponding to this heart rate. Slow baseline drift (caused by head movements) at frequencies below this range and high-frequency noise (caused by momentary facial expressions) at frequencies above this range are filtered out. After dual correction processing involving separation and filtering, the final output is a facial photoplethysmography (PPG) signal that reflects only the rhythm of the heart's pumping action.

[0035] For example, suppose that during a 10-second access control verification, the system extracts an initial signal. This signal contains two obvious anomalies: at the 3rd second, because someone walks by and blocks the ambient light, the initial signal exhibits a step fluctuation that causes the amplitude to drop by 40% instantaneously; at the 7th second, the user slightly frowns, generating an extremely high-frequency spike pulse that lasts only 0.1 seconds.

[0036] During calibration, the system first uses a blind source separation algorithm to remove the light step fluctuation at the 3-second mark from the green light channel data. The system sets the frequency threshold of the bandpass filter to 0.7 Hz to 3.0 Hz, which corresponds to the reasonable physiological range of human heart rate between 42 and 180 beats per minute. When the signal, after eliminating light interference, passes through this bandpass filter, the high-frequency spike of frowning that appears at the 7-second mark and lasts only 0.1 seconds is identified by the filter as non-physiological noise and directly filtered out; at the same time, the slow baseline drift with a frequency below 0.7 Hz caused by slight head shaking is also flattened.

[0037] Ultimately, the system outputs a clean rPPG signal that filters out all external interference and exhibits uniform periodic fluctuations between 0.7 and 3.0 Hz. By calculating the peak interval of this smooth waveform, the system can accurately identify the current user's heart rate as 75 beats per minute.

[0038] Optical response analysis of the fingertip optical reflection sequence was performed, specifically including: The system controls the fingertip optical acquisition device to emit detection light of a preset wavelength towards the fingertip area and acquires the reflection response data generated by the fingertip area in response to the detection light. When a user presses their finger on the optical acquisition area of ​​the access control terminal, the main control chip triggers the underlying LED light-emitting array to emit detection light of the preset wavelength towards the dermis layer of the fingertip. This preset wavelength is typically configured to be the green light band in the visible light spectrum, which has a very high absorption rate for hemoglobin containing oxygen in the blood. After the detection light penetrates the surface of the fingertip skin, some photons are absorbed by the blood in the subcutaneous blood vessels, while the unabsorbed photons are scattered and reflected, and received by the photodetector inside the acquisition device. The photodetector converts the received photon energy into an analog current, and the system's built-in analog-to-digital converter quantizes this analog current, outputting discrete digital values. These digital values ​​constitute the reflection response data used to characterize the current instantaneous light intensity.

[0039] The system extracts the characteristics of blood volume changes in the fingertip based on the light intensity changes corresponding to different sampling times. It then performs time-series processing on the continuously acquired optical response data to generate a fingertip photoplethysmography (PPG) signal. With the periodic diastolic and vasoconstriction of the heart, the blood volume in the fingertip microvessels undergoes periodic filling and depletion. When the heart contracts and the blood vessels are full of blood, the absorption of detection light by the blood reaches its maximum, resulting in the weakest light intensity reflected back to the photodetector; conversely, when the heart relaxes, the reflected light intensity is the strongest. The system uses an internal high-frequency timer to continuously read the above digital values ​​at a fixed sampling rate.

[0040] During continuous reading, due to uneven pressure from the user's fingers or slight muscle tremors, the raw continuous response data often contains high-frequency noise spikes and low-frequency pressure baseline drift. The system uses a sliding data window to perform time-series processing on this raw data. Specifically, the system sets a data window containing several continuous sampling points, calculates the arithmetic mean of all values ​​within the window as the smoothed output value at the center point, and then slides the window backward along the time axis, replacing the old data one by one and calculating the new result. This smoothing process eliminates irregular fluctuations caused by changes in the physical pressure of the fingers, ultimately connecting a series of discrete and smooth light intensity absorption feature points into a continuous fingertip photoplethysmography (FP-P) signal.

[0041] For example, suppose the access control system's fingertip optical sensor uses a green LED with a center wavelength of 525 nanometers as its light source, and the analog-to-digital converter has a quantization precision of 12 bits, meaning the output reflection response data ranges from 0 to 4095. The system's sampling frequency is set to 100 Hz, meaning it collects light intensity data 100 times per second. When a user presses their finger, the system acquires 200 consecutive discrete values ​​within the first 2 seconds. During cardiac systole, when subcutaneous blood is at its peak and absorbs the most green light, the photodetector receives very weak reflected light, and the output digital value drops to 1850. During cardiac diastole, when blood flow decreases, reflected light increases, and the output digital value rises to 2120.

[0042] Among these 200 continuously changing values, slight finger tremors caused some jerky spikes, such as 1805 and 2150. The system used a moving average window with a length of 5 sampling points for timing processing. When the window covered 5 consecutive data points containing the jerky spikes, the system automatically calculated the average of these 5 values, thus neutralizing the jerky spikes. Finally, the system output a continuous fluctuation curve composed of 200 smoothed digital values, which perfectly reproduced a pure fingertip PPG signal with an amplitude between 1850 and 2120 and a period of approximately 0.8 seconds, corresponding to a heart rate of 75 beats per minute.

[0043] Determining the timing relationships of physiological conduction specifically includes: The system extracts periodic features from the facial and fingertip photoplethysmography (rPPG) signals, respectively, and establishes a temporal correspondence between the two signals based on a unified time identifier. After extracting continuous facial and fingertip rPPG signals, the system first performs extreme value detection on these two signals in the time domain. Since each strong contraction of the heart pumps blood, a maximum blood volume is generated in the peripheral blood vessels, the system identifies and extracts the "systolic main peak" generated by heart contraction in these two signal waveforms, using it as the periodic feature in the pulse change characteristics. Next, using the unified time identifier generated in step S1 as the absolute time zero point, the system maps the main peaks of the facial signal and the fingertip signal onto the same time coordinate axis, accurately recording the absolute physical timestamp corresponding to the appearance of each main peak, thereby establishing a temporal correspondence between the two in the same spacetime.

[0044] Based on the order of occurrence of periodic changes in two pulse wave signals, the system determines the direction of pulse conduction between the facial and fingertip regions and assesses whether this direction conforms to the physiological time sequence of human blood circulation. Based on basic human anatomy, the physical distance that oxygenated blood pumped by the heart travels through the carotid artery to the facial capillaries is significantly shorter than the physical distance that travels through the subclavian and brachial arteries to the fingertip capillaries. Therefore, for the same living organism, the blood surge generated by the same heartbeat reaches the face first, and then the fingertips. The system determines the direction of pulse conduction by comparing the timestamps of the two main wave peaks on the same timeline. The system internally presets a physiological conduction time difference interval, with a lower limit strictly greater than zero and an upper limit set as the maximum reasonable delay in the conduction of a normal adult pulse wave to the extremities. If the system detects that the main peak of the facial signal appears earlier than the main peak of the fingertip signal in time, and the time difference between the two falls exactly within the preset physiological conduction time difference range, then it is determined that the current pulse conduction direction conforms to the real human blood circulation physiological time sequence relationship from near to far; otherwise, if the time difference between the two is zero or the fingertip appears earlier than the face, it is determined that it does not conform to the physiological time sequence relationship and is suspected of being a spoofing attack.

[0045] For example, the access control system has a pre-defined physiological conduction time difference range of 15 to 120 milliseconds, covering the cardiac pumping time difference characteristics of people with different heights and blood pressures. During an access control verification process, the system uses the moment of trigger signal capture as the absolute zero point. After extracting the peaks, the system finds that the first systolic peak of the facial rPPG signal appears at 800 milliseconds on the time axis; while the corresponding peak of the fingertip PPG signal appears at 845 milliseconds. The system calculates that the facial peak appears at 800 milliseconds earlier than the fingertip peak at 845 milliseconds, proving that the pulse conduction direction is face first, then fingertip. The system then calculates the peak occurrence time difference to be 45 milliseconds. Since 45 milliseconds falls within the pre-defined physiological conduction time difference range of 15 to 120 milliseconds, the system determines that this timing relationship perfectly matches the physiological characteristics of the real human body.

[0046] As a comparison of offense and defense, if an attacker uses a tablet to simultaneously play pre-recorded facial videos and fingertip pulse videos for a replay attack, the facial and fingertip pulse peaks extracted by the system will precisely appear at 800 milliseconds on the timeline, with a time difference of 0 milliseconds, since 0 milliseconds cannot fall into a preset range greater than zero. The system will immediately recognize that this physiological conduction direction violates the objective laws of physical distance, thus directly intercepting this spoofing attack.

[0047] Determining waveform consistency specifically includes: Periodic variation segments from facial and fingertip photoplethysmography (PPG) signals are acquired according to preset time windows. After continuously acquiring these two types of pulse wave signals, the system does not perform blind calculations on the infinitely long data stream; instead, it introduces a preset time window for data extraction. The duration of this preset time window is typically configured to include three to five complete heartbeat cycles. Using this time window, the main control chip simultaneously extracts a data slice from the facial pulse wave signal and a data slice from the fingertip pulse wave signal in the same time coordinate system as the periodic variation segments to be analyzed.

[0048] The system performs correlation analysis on the changing trends, periodic characteristics, and fluctuation patterns of two pulse wave signals within corresponding time windows. The waveform consistency is determined based on the degree of synchronization between the two signals to determine whether they originate from the same human physiological activity. In real physiological activity, due to the regulation of the autonomic nervous system and the influence of respiratory activity, the human heartbeat is not an absolutely uniform mechanical vibration, but rather exhibits microsecond-level acceleration or deceleration. This minute fluctuation pattern originates from the same heart and is precisely transmitted to the face and fingertips. The system performs correlation analysis on two extracted periodic change segments. Specifically, it extracts the signal amplitude values ​​of the two segments at each sampling time and performs normalization processing. The mean absolute difference algorithm is used to calculate the degree of synchronization between the two data sequences. The specific waveform consistency determination formula is as follows: ,in, It represents the degree of synchronization between two signals, i.e., the waveform consistency coefficient; This indicates the total number of sampling points captured within the preset time window; Indicates the first The amplitude value of the normalized facial photoplethysmography signal at each sampling time; Indicates the first The amplitude value of the fingertip photoplethysmography signal after normalization at each sampling time; It represents the absolute difference between the two at the same moment.

[0049] The result of the formula The closer the value is to 1, the more synchronized the changes of the two signals. The system has a preset waveform consistency threshold used for quantitative judgment. If the system calculates the degree of synchronization between the two signals... If the waveform consistency is greater than or equal to the preset threshold, it means that the face and fingertips are not only jumping, but even the occasional irregular tremors are completely synchronized, thus determining that the two signals share the same blood circulation center and originate from the same human physiological activity; conversely, if the calculated degree of synchronization change is lower than the threshold, it means that the two are just independent periodic vibrations, which is judged as a combined forgery attack.

[0050] For example, suppose the access control system has a preset time window of 4 seconds and a sampling rate of 100 Hz. Therefore, within each window, the system captures 400 facial pulse signal data points and 400 fingertip pulse signal data points. The system's preset waveform consistency threshold is set to 0.85, with a range of -1 to 1; the closer to 1, the higher the synchronization. In a real-world liveness detection test, the user took a slight breath at the 2-second mark, causing a very brief acceleration in their heart pumping rhythm, shortening the peak interval by 10 milliseconds. At this moment, both the captured facial pulse and fingertip pulse segments synchronously recorded this accelerated fluctuation. The system calculated the correlation between these two sequences containing 400 data points, obtaining a synchronization change coefficient of 0.93. Since 0.93 is greater than the preset threshold of 0.85, the system determines that the waveform consistency is extremely high, and both originate from the same living individual.

[0051] As a comparison for anti-spoofing, if an attacker holds the device in their left hand and plays a video of employee A's facial pulse (heart rate stable at 70 beats per minute), while simultaneously pressing the fingerprint module with their right hand and inputting employee B's severed finger or a recorded fake pulse (heart rate stable at 72 beats per minute), within a 4-second window, although both are real pulse waveforms, the microscopic fluctuations appear random and scattered due to the completely different neural regulation and breathing rhythms of employees A and B. The system calculates the synchronization change coefficient for these two sets of pieced data, finding it to be only 0.21. Since 0.21 is far below the preset threshold of 0.85, the system immediately determines the waveform inconsistency and successfully intercepts the pieced-together attack.

[0052] Before generating the liveness authentication identifier, there is also an abnormal authentication status judgment step, which includes: The system monitors the temporal correspondence and waveform changes between facial and fingertip photoplethysmography (PPG) signals in real time. In the main control chip's memory buffer, the system simultaneously receives and aligns these two pulse wave signals, while simultaneously launching a parallel monitoring thread. This monitoring thread does not involve complex identity model inference; instead, it compares the relative positions of the peak timestamps in the two signals in real time, as well as the statistical correlation score between the two signal waveforms.

[0053] When there is no time relationship between two pulse wave signals that conforms to the physiological conduction law of the human body, or when the changing trend of the two pulse wave signals does not meet the preset consistency condition, it is determined that there is an abnormal state in the current authentication process, and the generation of the liveness authentication mark is prevented.

[0054] The system categorizes abnormal states into two typical trigger conditions, and the circuit breaker mechanism is immediately triggered if either condition is met. The first type of abnormality is a time relationship that violates physiological common sense: the system presets a time difference range for physiological pulse conduction. If the monitoring thread finds that the absolute time difference between the facial and fingertip peaks is zero, it means the two signals are absolutely synchronized in time, usually a phenomenon caused by playing recorded video on a screen; or the facial peak appears later than the fingertip peak, exhibiting reverse conduction characteristics; or the time difference between the two greatly exceeds the upper limit of the preset range, proving that they are independent pulsation sources with no connection. All three situations are judged by the system as lacking a time relationship consistent with the physiological conduction laws of the human body. The second type of abnormality is a disjointed trend: the system sets an extremely high lower threshold for waveform correlation. Even if an attacker simulates a reasonable conduction time difference using highly sophisticated instruments, the system cannot simulate the subtle irregular heart rhythm fluctuations caused by respiration and the autonomic nervous system, resulting in the monitoring thread calculating a correlation score between the two signal segments below this lower threshold. In this case, the system determines that it does not meet the preset consistency condition. Once the abnormal state is determined, the system's underlying layer directly clears all video streams and optical sequence data in the current memory buffer and reports the abnormal alarm log to the IoT gateway. At the same time, it completely blocks the generation process of the liveness authentication identifier, preventing the system from proceeding to the subsequent step S5.

[0055] Joint authentication based on identity features specifically includes: Facial structural features are extracted from the liveness-authenticated facial video stream, and fingerprint texture features are extracted from the fingertip image data corresponding to the fingertip optical reflection sequence. After the system successfully generates an absolute liveness authentication identifier in step S4, the main control chip is activated and begins the computationally intensive identity feature extraction operation. For facial data, the system extracts the clearest frame of the baseline image from the cached facial video stream, uses a convolutional neural network to extract facial structural features including the locations of key bones and muscles such as the corners of the eyes, the tip of the nose, and the cheekbones, and converts it into a fixed-dimensional facial feature vector. For fingertip data, since the fingertip optical reflection sequence acquired in step S1 is essentially a series of internal light intensity images of the finger, the system recovers a high-contrast two-dimensional fingerprint surface texture image from it and extracts microscopic details such as ridge direction, endpoints, and bifurcation points, converting them into a fingerprint feature vector.

[0056] Facial structural features and fingerprint texture features are matched against pre-stored authorized identity information in the access control system. The system database pre-stores authorized identity template information entered by company employees, also encrypted and stored in the form of feature vectors. The main control chip calculates the spatial distance between the currently extracted facial feature vector and the pre-stored facial template vector in the database, converting it into a percentage-based facial matching similarity score. Simultaneously, it calculates the spatial distance between the currently extracted fingerprint feature vector and the pre-stored fingerprint template vector, converting it into a percentage-based fingerprint matching similarity score. These two scores directly reflect the degree to which the currently collected data matches the identity of the user registered in the database.

[0057] When both facial structural features and fingerprint texture features meet the corresponding authentication conditions, an access control command is generated. The system sets independent authentication thresholds for both face and fingerprint. Only when the facial matching similarity score is greater than or equal to the preset facial authentication threshold, and the fingerprint matching similarity score is greater than or equal to the preset fingerprint authentication threshold, does the system determine that both of the user's identities have been verified legitimately. At this point, the main control chip sends a level transition signal to the physical actuator of the access control system, generating a data packet containing an unlocking command to allow the user to pass. If either score fails to reach the corresponding threshold, the system will remain locked.

[0058] For example, suppose a secure access control system in a confidential computer room has a high security level, with a preset facial recognition threshold of 85 points and a preset fingerprint recognition threshold of 80 points. During a routine access process, authorized employee A clocks in. After confirming that A is a live person, the system extracts A's facial features and compares them with a template of employee A in the database, calculating a facial similarity score of 92 points; simultaneously, it extracts A's fingerprint features and compares them, calculating a fingerprint similarity score of 88 points. The system determines that 92 points is greater than the facial recognition threshold, and 88 points is greater than the fingerprint recognition threshold. Both features meet the corresponding authentication conditions, and the system then sends a high-level command to the electronic door lock, opening the door.

[0059] In contrast, if an external person, B, finds employee A's access card and somehow passes the liveness detection at the access gate, the system extracts external person B's facial features for comparison and finds that the facial similarity score is only 35 points. Even if external person B somehow provides a feature that would give a fingerprint score of 82 points, the system still refuses to generate an opening command because the facial condition is not met, thus ensuring the security of the physical space.

[0060] It should be noted that all formulas in this manual are calculated by removing dimensions and taking their numerical values. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0061] The above description is merely a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for users of ordinary technical skills, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A smart access control method based on the Internet of Things, characterized in that, include: S1. Obtain the authentication trigger signal and collect a facial video stream and fingertip optical reflection sequence containing a unified time identifier based on the global clock synchronization mechanism; S2. Extract physiological signals from the facial region in the facial video stream to generate facial photoplethysmography (PPG) signals, and perform optical response analysis on the fingertip optical reflection sequence to generate fingertip PPG signals. S3. Based on the unified time identifier, the facial photoplethysmography (PPG) wave signal and the fingertip PPG wave signal are time-synchronized, the corresponding pulse change features are extracted, and the physiological conduction timing relationship and waveform consistency between the two are determined. S4. Determine whether the preset liveness authentication conditions are met based on the physiological conduction timing relationship and waveform consistency. If the conditions are met, generate a liveness authentication identifier. S5. In response to the liveness authentication identifier, perform joint authentication on the identity features in the facial video stream and the fingertip optical reflection sequence, and generate an access control command when the authentication is successful.

2. The intelligent access control method based on the Internet of Things as described in claim 1, characterized in that, The process in S1, which involves acquiring a facial video stream and fingertip optical reflection sequence containing a unified time identifier based on a global clock synchronization mechanism, specifically includes: Receive the authentication trigger signal sent by the access control terminal and extract the basic time information corresponding to the authentication trigger signal; Based on the basic time information and the device response delay information corresponding to the facial image acquisition device and the fingertip optical acquisition device, the time reference of different acquisition devices is calibrated to generate a unified time identifier. Based on the unified time identifier, the facial image acquisition device is controlled to acquire continuous facial video frames, and the fingertip optical acquisition device is simultaneously controlled to acquire optical reflection data within the corresponding time period.

3. The intelligent access control method based on the Internet of Things as described in claim 1, characterized in that, S2 involves extracting physiological signals from the facial region in the facial video stream, specifically including: Face detection and feature localization are performed on consecutive image frames in the facial video stream to obtain positional change information of key facial regions; Based on the key facial regions, an effective area for physiological signal acquisition is determined, including the forehead region and the cheek region; Color change information corresponding to multiple consecutive image frames within the effective area is extracted, and the color change information is converted into a time series to generate a facial photoplethysmography (PPG) signal reflecting changes in facial blood volume.

4. The intelligent access control method based on the Internet of Things as described in claim 1, characterized in that, The step of generating a facial photoplethysmography (PPG) signal in S2 includes the steps of generating an initial signal and suppressing interference, specifically including: Acquire interference information caused by changes in ambient light, facial posture, and facial expression during the facial video stream acquisition process; The initial signal is corrected based on the interference information to reduce the influence of non-physiological factors on the signal change trend, thereby obtaining the facial photoplethysmography signal.

5. The intelligent access control method based on the Internet of Things as described in claim 1, characterized in that, The optical response analysis of the fingertip optical reflection sequence in step S2 specifically includes: The fingertip optical acquisition device is controlled to emit detection light of a preset wavelength towards the fingertip area, and the reflection response data of the fingertip area to the detection light is acquired. Based on the light intensity change information corresponding to different sampling times, the characteristics of blood volume change in the fingertip are extracted, and the optical response data obtained by continuous acquisition are processed in a time sequence to generate the fingertip photoplethysmography (FP-P) signal.

6. The intelligent access control method based on the Internet of Things as described in claim 1, characterized in that, Determining the physiological conduction timing relationship in step S3 specifically includes: Periodic features in the corresponding pulse change features are extracted from the facial photoplethysmography (PPG) signal and the fingertip PPG signal, respectively, and a time correspondence between the two pulse signals is established based on the unified time identifier. Based on the order of appearance of the periodic change characteristics in the two pulse wave signals, the pulse conduction direction between the facial region and the fingertip region is determined, and it is judged whether the pulse conduction direction conforms to the physiological time sequence relationship corresponding to the human blood circulation process.

7. The intelligent access control method based on the Internet of Things as described in claim 1, characterized in that, Determining the waveform consistency in step S3 specifically includes: According to a preset time window, periodic change segments in the facial photoplethysmography (PPG) signal and the fingertip PPG signal are obtained respectively. The correlation analysis is performed on the changing trends, periodic characteristics and fluctuation patterns of the two pulse wave signals within the corresponding time window. The consistency of the waveforms is determined based on the degree of synchronous change between the two signals, so as to determine whether the two pulse wave signals originate from the same human physiological activity.

8. The intelligent access control method based on the Internet of Things as described in claim 1, characterized in that, Before generating the liveness authentication identifier in step S4, an abnormal authentication status judgment step is also included, specifically including: Real-time detection of the temporal correspondence and waveform changes between the facial photoplethysmography (PPG) signal and the fingertip PPG signal; When there is no time relationship between two pulse wave signals that conforms to the physiological conduction law of the human body, or when the changing trend of the two pulse wave signals does not meet the preset consistency condition, it is determined that there is an abnormal state in the current authentication process, and the generation of the liveness authentication identifier is prevented.

9. The intelligent access control method based on the Internet of Things as described in claim 1, characterized in that, The identity feature-based joint authentication in S5 specifically includes: Facial structural feature information is extracted from the facial video stream that has passed liveness authentication, and fingerprint texture feature information is extracted from the fingertip image data corresponding to the fingertip optical reflection sequence; The facial structure feature information and the fingerprint texture feature information are respectively matched with the authorized identity information pre-stored in the access control system; When both the facial structure feature information and the fingerprint texture feature information meet the corresponding authentication conditions, the access control command is generated.

10. An IoT-based intelligent access control system, implemented by any one of claims 1 to 9, characterized in that, include: The synchronous acquisition module acquires the authentication trigger signal and acquires a facial video stream and fingertip optical reflection sequence containing a unified time identifier based on a global clock synchronization mechanism; The signal extraction module extracts physiological signals from the facial region in the facial video stream to generate facial photoplethysmography (PPG) signals, and performs optical response analysis on the fingertip optical reflection sequence to generate fingertip PPG signals. The timing alignment module performs timing synchronization of the facial photoplethysmography (PPG) wave signal and the fingertip PPG wave signal based on the unified time identifier, extracts the corresponding pulse change features, and determines the physiological conduction timing relationship and waveform consistency between the two. The liveness determination module determines whether the preset liveness authentication conditions are met based on the physiological conduction timing relationship and waveform consistency. When the conditions are met, a liveness authentication identifier is generated. The access control module, in response to the liveness authentication identifier, performs joint authentication of the identity features in the facial video stream and the fingertip optical reflection sequence, and generates an access control command when the authentication is successful.

Images