Smart home security system based on Internet of Things

Abstract

The invention belongs to the technical field of smart home security, and particularly discloses a smart home security system based on the Internet of Things, and the system comprises an image recognition module which is used for collecting a face image, extracting face features, comparing the face features with a pre-stored white list, and judging whether a person belongs to a person in the white list or not; the dynamic connection module is used for setting a communication connection range as a first range or expanding the communication connection range as a second range when detecting that a person not on the white list exists; the equipment signal acquisition module is used for scanning and acquiring wireless identification signals of surrounding equipment in a communication connection range and extracting multi-dimensional information of the equipment so as to construct an equipment soft fingerprint; the identity binding module is used for performing time-space synchronous matching on the face features and the equipment soft fingerprints to generate a binding relationship between the face features and the equipment soft fingerprints; the method has the advantages that the recognition precision and tracking capability of the intelligent door lock are improved, and the endurance and security efficiency are both considered.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart home security, in particular to a smart home security system based on Internet of Things. BACKGROUND

[0002] The existing smart door lock usually adopts two independent authentication links of face recognition and fingerprint recognition: a low-resolution fixed-focus camera is responsible for capturing a close-up face image and performing a light neural network comparison locally, and a capacitive or semiconductor fingerprint sensor realizes touch confirmation through a fingerprint hash match; one-time judgment of each can trigger the unlocking. Due to the limitations of endurance and cost, it is ensured that the daily whitelist users can be unlocked within one meter without sensing, and the battery consumption is compressed. However, under this architecture, the camera resolution is not enough to capture sufficient facial details at one meter, and the fingerprint module relies on the user to actively touch. When a stranger stays or observes in front of the door for a short time, the system cannot accurately retain the high-quality features of his face, nor actively expand the Bluetooth listening range to collect his feature signals, resulting in only scattered or blurred pictures in the door lock log without associated feature identifiers.

[0003] To this end, a smart home security system based on Internet of Things is proposed to solve the above problems. SUMMARY

[0004] The present application aims to provide a smart home security system based on Internet of Things to solve or improve the above technical problems that the existing smart door lock is difficult to obtain relatively rich identifiable features of a stranger without active cooperation, resulting in ineffective marking and tracking of the stranger's identity.

[0005] Therefore, the first aspect of the present application is to provide a smart home security system based on Internet of Things.

[0006] The second aspect of the present application is to provide a method.

[0007] The first aspect of the present application provides an intelligent home security system based on Internet of Things, comprising: an image recognition module, configured to collect a face image of a person in front of a door lock and extract a face feature, compare the face feature with a pre-stored whitelist, and determine whether the person belongs to a whitelisted person; a dynamic connection module connected with the image recognition module, configured to set a communication connection range as a first range when the image recognition module detects that only a whitelisted person is in front of the door lock, or expand the communication connection range to a second range when the image recognition module detects that a non-whitelisted person is present; a device signal acquisition module connected with the dynamic connection module, configured to scan and acquire a wireless identification signal of a surrounding device within the communication connection range, extract multi-dimensional information of the device, and construct a device soft fingerprint; and an identity binding module connected with the image recognition module and the device signal acquisition module respectively, configured to perform space-time synchronous matching on the face feature and the device soft fingerprint, generate a binding relationship between the face feature and the device soft fingerprint, and store a basis for personnel identity recognition.

[0008] In any of the above technical solutions, the image recognition module comprises: a face detection unit, configured to detect faces in a region in front of the door lock in real time in a wake-up state, and determine positions and quantities of the faces; a feature extraction unit connected with the face detection unit, configured to perform feature embedding on the detected faces, and extract face feature vectors; and an identity comparison unit connected with the feature extraction unit, configured to perform similarity calculation on the face feature vectors and a pre-stored whitelist, and determine whether the faces correspond to whitelisted persons.

[0009] In any of the above technical solutions, the communication connection range of the dynamic connection module further comprises a third range, the face detection unit is started when a device appears in the third range, and the face detection unit is in a wake-up state.

[0010] In any of the above technical solutions, a space size of the third range is smaller than that of the first range, and when the face detection unit is in the wake-up state, the communication connection range changes from the third range to the first range or the second range.

[0011] In any of the above technical solutions, the dynamic connection module comprises: a range decision unit, configured to select the first range or the second range according to a result determined by the image recognition module; and a communication parameter adjustment unit connected with the range decision unit, configured to set power parameters and broadcast periods of communication connection in the first range, the second range and the third range, respectively.

[0012] In any of the technical solutions above, the device signal acquisition module comprises: a signal scanning unit, configured to continuously scan signals broadcast by nearby devices within a communication connection range, and to acquire multi-dimensional information of the devices; and a soft fingerprint generation unit, connected to the signal scanning unit, configured to construct and store the device soft fingerprint according to the multi-dimensional information containing device identity information.

[0013] In any of the technical solutions above, the device signal acquisition module further comprises: an identity injection interaction unit, configured to actively broadcast an identity handshake request signal, and to monitor whether surrounding devices send a response; and an identity authentication response unit, connected to the identity injection interaction unit, configured to receive binding identification information sent by a responding device, and to send the binding identification information to the soft fingerprint generation unit as the device identity information.

[0014] In any of the technical solutions above, the identity binding module comprises: a space-time synchronization unit, configured to perform space-time matching on the face feature and the device soft fingerprint, and to determine the relevance of the face feature and the device soft fingerprint; and a binding confirmation unit, connected to the space-time synchronization unit, configured to confirm a binding relationship based on a relevance determination result, and to form a binding record of the face feature and the device soft fingerprint.

[0015] In any of the technical solutions above, the identity binding module further comprises: a risk marking unit, connected to the binding confirmation unit, configured to mark a corresponding person as a potential risk person and to generate an alarm information when the binding confirmation unit determines that a binding confidence of the face feature and the device soft fingerprint is lower than a preset threshold; and a binding database unit, connected to the binding confirmation unit, configured to store and manage the binding record.

[0016] The second aspect of the application provides a method, comprising the following steps: acquiring a face image of a person in front of a door lock, extracting a face feature, and comparing the face feature with a white list; adjusting the communication connection range to the first range or the second range according to the comparison result; scanning wireless communication signals sent by surrounding devices within the adjusted communication connection range, extracting the multi-dimensional information, and generating the device soft fingerprint; performing space-time matching on the device soft fingerprint and the corresponding face feature within a same time window to obtain a binding relationship of the face feature and the device soft fingerprint, and storing the binding relationship as a basis for personnel identity recognition.

[0017] Compared with the prior art, the application has the following beneficial effects: By combining image recognition, communication range dynamic adjustment, wireless device signal collection and face-device soft fingerprint space-time binding in the intelligent door lock, multi-dimensional enhancement of home entry safety is realized: the image recognition module determines the personnel identity in advance, significantly improving the authentication speed and accuracy of the white list personnel; the dynamic connection module flexibly adjusts the communication range according to the personnel category, reduces the communication range and power consumption when only the white list members appear, and expands the communication range and senses the risk in time when unknown personnel appear, which balances the endurance performance and safety monitoring effect of the door lock; the device signal collection module can actively collect multi-dimensional signals of unknown devices within the extended communication range and construct a soft fingerprint, which greatly improves the recognition and tracking ability of strangers and their electronic devices; through the identity binding module, the face features and device soft fingerprint are accurately matched in space and time, and the binding record is stored, which effectively makes up for the problems that traditional face recognition cannot capture device information and low-resolution cameras cannot accurately identify strangers, so as to realize efficient marking, timely warning and reliable tracing of subsequent events of non-white list personnel, and overall improve the intelligence, accuracy and safety of the home intelligent security system.

[0018] Additional aspects and advantages of embodiments according to the present application will become apparent from the following description with reference to the drawings. BRIEF DESCRIPTION OF DRAWINGS

[0019] The above and / or additional aspects and advantages of the present application will become apparent and more readily appreciated from the following description, taken in conjunction with the accompanying drawings, in which:

[0020] Figure 1 is a system logic block diagram of the present application;

[0021] Figure 2 is a system main flowchart of the present application;

[0022] Figure 3 is a dynamic connection module double-state flowchart of the present application;

[0023] Figure 4 is a device signal collection and binding flowchart of the present application;

[0024] Figure 5 is a risk disposal flowchart of the present application;

[0025] Figure 6 is a multi-module linkage timing diagram of the present application;

[0026] Figure 7 is a method step flowchart of the present application. DETAILED DESCRIPTION

[0027] In order to make the above objects, features and advantages of the present application more clearly understood, the following further describes the present application with reference to the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0028] In the following description, a large number of specific details are set forth in order to facilitate a thorough understanding of the present application, but the present application can also be implemented in other manners different from those described herein, and therefore the scope of protection of the present application is not limited to the specific embodiments disclosed below.

[0029] Please refer to Figures 1-7 , the following describes some embodiments of the present application, an intelligent home security system based on Internet of Things.

[0030] Embodiments of the first aspect of the present application propose an intelligent home security system based on Internet of Things. In some embodiments of the present application, as shown in Figures 1-6 , the intelligent home security system comprises: An image recognition module is configured to collect a face image of a person in front of a door lock and extract a face feature, compare the face feature with a pre-stored white list, and determine whether the person belongs to a white list person.

[0031] A dynamic connection module is connected to the image recognition module and is configured to set a communication connection range as a first range when the image recognition module detects that only a white list person is present in front of the door lock, or expand the communication connection range to a second range when the image recognition module detects that a non-white list person is present.

[0032] A device signal acquisition module is connected to the dynamic connection module and is configured to scan and acquire a wireless identification signal of a surrounding device within the communication connection range, extract multi-dimensional information of the device, and construct a device soft fingerprint.

[0033] An identity binding module is connected to the image recognition module and the device signal acquisition module, respectively, and is configured to perform time-space synchronous matching of the face feature and the device soft fingerprint, generate a binding relationship between the face feature and the device soft fingerprint, and store the binding relationship as a basis for personnel identity recognition.

[0034] The intelligent home security system based on Internet of Things provided by the present application is usually installed on a smart door lock or a position near the smart door lock, and is composed of a camera, an image processing chip and an image algorithm component. When a person is detected to approach the front of the door lock, the camera is automatically started to collect a face image of the person. The camera device can be an embedded infrared camera or a wide-angle visible light camera, supporting dual-mode recognition in daytime and nighttime. After the image is collected, the system quickly locates a face region in the image through an integrated face detection algorithm, and extracts key feature information of the face using a lightweight embedded algorithm to generate a corresponding high-dimensional face feature vector.

[0035] Subsequently, the face feature vector will be sent into the local identity comparison process, and the similarity comparison is carried out with the pre-stored whitelist data in the system. The whitelist can be recorded by the user in advance through the mobile phone App or the management background, including the face feature vectors of authorized family members, regular visitors and other personnel. The comparison process uses Euclidean distance, cosine similarity or learning-based classification network for measurement. When the similarity is higher than the set threshold, the system determines that the person is a whitelist user, otherwise it is marked as a non-whitelist or unknown person.

[0036] Specifically, when the homeowner carries a mobile phone close to the entrance door, the face features are collected and processed by the camera and recognized as a whitelist member. At this time, the image recognition module will return the authenticated personnel flag, which provides the basis for decision-making for subsequent communication strategy adjustment and unlocking authorization. On the contrary, if the camera captures an unregistered person (such as a courier, a stranger, etc.), the system will record his face features and report it as a non-whitelist person, triggering the next step of dynamic connection adjustment and risk recording logic.

[0037] The dynamic connection module is connected with the image recognition module, which is used to dynamically adjust the parameters of wireless communication connection according to the recognition result, so as to ensure the endurance performance of the door lock while improving the monitoring ability of potential abnormal personnel. Specifically, the dynamic connection module adjusts the transmission power, scanning frequency, broadcast period and other parameters of short-range communication technologies such as Bluetooth, Wi-Fi, UWB or NFC by controlling the switching of communication connection range, thereby changing the spatial coverage range that the device can sense.

[0038] In the case where the system detects that only whitelist personnel exist in front of the door lock, the dynamic connection module limits the communication connection range to the first range, i.e. the preset communication parameter configuration with smaller coverage distance and lower power consumption. For example, the Bluetooth transmission power can be set to -20dBm, and the broadcast period can be set to more than 1 second, only maintaining a short-distance and intermittent device listening state. In this mode, the focus is on the unlocking intention of the whitelist personnel, and the low-frequency identification interaction is carried out on the devices such as smart phones and wearable devices carried by the whitelist personnel, so as to significantly reduce the power consumption of the communication module and effectively prolong the battery life cycle without affecting the user experience.

[0039] Correspondingly, when the image recognition module detects the presence of a non-whitelisted person in front of the door lock, the dynamic connection module will immediately adjust the communication connection range to a second range, i.e., a larger space, higher power, and higher frequency scanning mode. In this mode, the system increases the transmission power of Bluetooth or Wi-Fi modules to the maximum value, such as 0~4dBm, shortens the broadcast period to 100ms, and continuously scans the space around the door lock at a high frequency to obtain as much wireless identification signal as possible from the devices carried by strangers, such as MAC address, SSID, device model, etc. Such data will be used by subsequent modules to construct a soft fingerprint to assist in identity labeling and event tracing.

[0040] Specifically, when an unregistered stranger stands in front of the door to observe or stay, the image recognition module does not match its facial features in the whitelist, and immediately notifies the dynamic connection module to enter the alert mode, expanding the communication range to a 5~8 meter radius range and continuously obtaining all detectable device broadcast data in the surrounding area. At the same time, a directional antenna or TOA (Time of Arrival) estimation method can also be enabled to determine the direction of the device signal, thereby preliminarily establishing the spatio-temporal correspondence between the face image and the wireless device. If the detected device repeatedly appears in multiple events, the system can also record it as a suspicious device and suggest the resident to check or alarm.

[0041] The device signal acquisition module is connected with the dynamic connection module, which is used to actively or passively scan the wireless identification signals sent by the surrounding devices within the communication connection range, and to analyze and extract the features of the obtained signals, finally generating a device soft fingerprint that can be used for identity identification and behavior trajectory tracking. In the system, it is responsible for sensing the existence of devices, building person-device associations, especially when strangers do not directly touch the door lock, it can still achieve remote identification and subsequent tracking with the help of their personal electronic device information. Specifically, the device signal acquisition module continuously operates within the communication connection range set by the dynamic connection module, listens to the short-range identification signals sent by various wireless devices with broadcasting capabilities, which can come from Bluetooth, Wi-Fi probe request, NFC tag, UWB ultra-wideband response signal, or other intelligent terminals with passive transmission capabilities. The scanning unit supports multi-protocol parallel listening and can cross-platform identify the broadcasting behavior of devices from different manufacturers, and real-time extract the parameters with differences and identifiability as feature data sources.

[0042] Specifically, a stranger visitor appears in front of the door lock, who does not touch the door lock actively, but the Wi-Fi probe request automatically sent by his mobile phone is captured by the system. The request contains multiple dimensions of content such as device type, manufacturer feature code, MAC address (which can be randomized), broadcast period, RSSI signal strength, etc. The device signal collection module receives it and sends it to the soft fingerprint generation process.

[0043] Specifically, the above-mentioned collected raw data constructs a multi-dimensional feature vector. The multi-dimensional information includes but is not limited to the following aspects: Broadcast signal type: Identify whether it is a Wi-Fi, Bluetooth, NFC or UWB signal, which helps to determine the device type; Signal transmission frequency band: such as 2.4GHz, 5GHz, 6GHz, etc., used to preliminarily infer the communication chip model or protocol version; Broadcast field content: such as the device name, service UUID, and manufacturer identification code contained in the Bluetooth broadcast; Signal strength change curve (RSSI trajectory): The strength-time change image constructed after receiving multiple periodic signals in succession; Signal appearance frequency and period: Reflects the device broadcast interval, which helps to distinguish between system settings (such as Android / iOS differences); Signal response delay and interaction mode: such as the response delay of the device to the scanning request and whether it supports handshake response; Signal source relative direction estimation: Preliminary judgment of the relative position of the device and the door lock based on multi-antenna RSSI or UWB ranging technology; MAC address mode: Whether to use random MAC (RPA), address replacement frequency, etc., reflecting the privacy setting tendency; Broadcast content change trend: Used to determine whether it is multiple disguises of the same device.

[0044] All the above features will be combined into a soft fingerprint vector with high recognition ability. The entire vector constitutes the unique signal behavior exhibited by a certain device in a specific time period and specific environment. After hash, encryption or fuzzy matching processing of the vector, it can be used in subsequent modules to bind space-time with image recognition results, thereby establishing the correspondence between the face image of a stranger and his device. Further, the system supports clustering or trajectory analysis of soft fingerprints generated by the same device at different time periods and different door appearances. When a device appears in front of the door lock of the same community at different times, it can be automatically marked as a frequent unknown device, prompting the manager to pay attention to whether there is a risk of stepping on or abnormal wandering.

[0045] The identity binding module is connected with the image recognition module and the device signal acquisition module respectively, and is used for performing time and space synchronization comparison on the face feature information extracted by the former and the device soft fingerprint information constructed by the latter, determining whether the two are derived from the same target individual in space, and judging whether the two types of information belong to the same event period in time, so as to establish a reliable person-device binding relationship and generate a corresponding identity binding record as an important basis for subsequent identification, tracking and risk analysis. Specifically, when the system detects the appearance of a person in front of the door and successfully extracts the face image feature vector thereof, the identity binding module obtains the device soft fingerprint set recorded in the device signal acquisition module at this moment. At this time, the system opens a binding time window, and the length of the time window can be adjusted according to the actual deployment strategy, for example, 2 seconds or 5 seconds, which is used to match the device broadcast behavior occurring almost simultaneously with the image event. At the same time, the system also calculates the spatial consistency index between the face appearance direction and the signal source direction by referring to the horizontal position of the face in the image plane in the current video frame and the relative receiving angle or RSSI strength gradient of the device signal.

[0046] Specifically, when the image recognition module detects the face feature of a stranger in front of the door lock and records the face vector feature , the device signal acquisition module captures two device signals and at the same time, wherein comes from the left side within 1 meter, and the signal strength is stable, comes from the area behind the door but the signal is intermittent. The identity binding module performs priority binding on and because the time sequence consistency (Δt < threshold), spatial direction consistency Δθ < threshold and signal behavior feature matching degree such as broadcast frequency and response time delay of the two are highly correlated. Once the identity binding module determines that the comprehensive score of a certain face feature and a certain device soft fingerprint exceeds the set threshold Sthresh, the system will confirm that the binding is established, and call the binding confirmation submodule to generate a formal identity binding record, including: face feature code, face image original screenshot, device soft fingerprint vector, binding timestamp, relative spatial position, binding confidence score and other information. The record will be stored in the binding database, and whether it is pushed to the user end for prompting or marking will be determined according to the risk strategy setting.

[0047] In addition, in order to further improve the recognition ability and long-term tracking effect of the system, the identity binding module supports continuous learning of new binding objects. Specifically, when a certain device is stably bound to a certain face feature for multiple times, the system will automatically mark the binding relationship as a trusted binding, and the next time it appears, it can be directly judged as a known object. On the contrary, if a certain device is bound to different faces in front of multiple door locks, the system can mark it as a suspicious device, triggering further image comparison or alarm mechanism.

[0048] In summary, the present application not only makes up for the three major defects of the prior art, i.e. difficulty in marking strangers by independent biological recognition, insufficient sampling by low-resolution cameras, and missing suspicious devices by fixed low-power Bluetooth, but also realizes a closed loop of identity determination-communication strategy-device awareness-risk confirmation in the same system: in a white list scenario, the dynamic connection module automatically converges the communication radius, significantly reducing wireless power consumption and prolonging battery life; when a stranger appears, the communication range is instantaneously enlarged, and the device signal acquisition module can capture wireless identification signals such as mobile phones / Wi-Fi / UWB in a larger space and generate multi-dimensional soft fingerprints; the identity binding module accurately matches these soft fingerprints with the corresponding face features within a microsecond time window, forming a traceable person-machine binding record. As a result, the system not only maintains the fast and unobtrusive unlocking experience of registered users, but also significantly improves the long-distance awareness, trajectory tracking and risk warning capabilities for outsiders, achieving simultaneous improvement of battery life optimization, security protection and evidence reliability.

[0049] In any of the above embodiments, the image recognition module comprises: a face detection unit for detecting faces in the area in front of the door lock in real time in the wake-up state, determining the position and number of faces.

[0050] a feature extraction unit connected to the face detection unit, configured to perform feature embedding on the detected faces and extract face feature vectors.

[0051] an identity comparison unit connected to the feature extraction unit, configured to perform similarity calculation between the face feature vectors and the pre-stored white list, and determine whether the corresponding personnel of the face is a white list personnel.

[0052] In this embodiment, when the door lock acceleration sensor or infrared distance sensor triggers a person approaching event, the system first wakes up the camera and starts the face detection unit, which uses the lightweight RetinaFace-Lite network to scan video frames in real time at ≈25fps, outputting the coordinates, confidence, and number of rectangular frames for each face. Next, the feature extraction unit performs 112x112 normalization preprocessing on the face cropped from each frame and calls MobileFaceNet-INT8 for embedding inference, obtaining an average of 30ms for a 128-dimensional face feature vector. Finally, the identity comparison unit searches for the nearest neighbor in the local whitelist feature library with a typical size of ≤50 people using cosine similarity. If the maximum similarity is ≥0.75, it is determined to be a whitelist person, otherwise it is marked as a non-whitelist or unknown person, and the results are sent to the dynamic connection module.

[0053] For example, a single face is detected with a similarity of 0.88, immediately determined to be a whitelist, the dynamic connection module maintains a low power -20dBm Bluetooth, directly pushes the lock open and wakes up the motor; the next morning, the courier visits, the camera recognizes a strange face with a confidence of 0.63, the system determines it to be a non-whitelist, triggering the communication range to upgrade from 1m to 8m, then the device signal acquisition module detects its phone BLE broadcast and generates a soft fingerprint, the identity binding module binds the fingerprint with the courier's face in the same time window, and records it as a stranger visitor-soft fingerprint IDA3BC…, for later backtracking or quick comparison when visiting again.

[0054] In any of the above embodiments, the communication connection range of the dynamic connection module also includes a third range, and when a device appears in the third range, the face detection unit is started and the face detection unit is in an awake state.

[0055] In this embodiment, when the system is waiting for a long time, the dynamic connection module maintains a very low power listening in the third range with a radius of about 30-50cm, a Bluetooth transmission power of -30dBm, and an advertising interval of ≥3s, only passively sweeping the short-range wireless communication signals close to the door lock; Once any device's short-range wireless communication signal including BLE, NFC or UWB handshake pulse is captured in the third range, an awake instruction is immediately issued to power on the camera and start the face detection unit, and the communication parameters are instantaneously switched to the normal power configuration of the first range or the second range. Subsequently, the image recognition module performs real-time face detection and whitelist comparison, and the dynamic connection module determines whether to maintain low power (whitelist determination) or enter high alert extended scanning (stranger determination) according to the comparison result, thereby realizing the intelligent wake-up and risk screening closed loop of device first and image later under the premise of ensuring the endurance.

[0056] Exemplarily, when the Bluetooth-enabled mobile phone is close to the door lock with the key package, the third range detects the BLE extended advertising packet sent by the mobile phone, and the camera is immediately triggered to wake up; the face detection unit identifies the face of the resident as a white list within 600 ms, the system maintains the first range low-power communication and completes the unlocking, and the entire process consumes less than 2 mAh of power. Conversely, during the evening rush hour, an unknown mobile phone attempts to pair with the door lock at a distance of 40 cm, and the third range triggers the camera to wake up again, but the camera does not detect any white list face and captures a stranger's face image. The dynamic connection module immediately increases the communication power to the second range radius of 5-8 m and continuously scans, records the soft fingerprint of the mobile phone, and pushes a suspicious proximity warning to the homeowner, achieving rapid identification and security response to potential tailing or detection behavior.

[0057] In any of the above embodiments, the third range has a smaller space size than the first range, and when the face detection unit is in the wake-up state, the communication connection range changes from the third range to the first range or the second range.

[0058] In this embodiment, in the multi-level communication strategy, the third range is designed as the minimum, lowest-power close-in monitoring zone with a radius of about 30-50 cm, and only maintains ultra-low power scanning when the camera is in sleep mode; Its physical coverage is significantly smaller than the first range of about 1 m for normal unlocking and the second range of about 5-8 m for alert. When any wireless identification signal is detected in the third range and the camera is triggered to wake up, the image recognition module starts to work. If the face detection unit confirms that the visitor belongs to the white list, the dynamic connection module immediately raises the communication parameters from the third range to the first range, maintaining reliable short-range interaction between the resident's mobile phone and the door lock while keeping low power consumption; If the detection result shows that there are non-white list personnel, the communication connection range is directly upgraded from the third range to the second range, and high-power, short-period continuous scanning is started to maximize the capture of stranger device signals and hand over to the subsequent soft fingerprint and risk analysis process.

[0059] Exemplarily, the mobile phone establishes a BLE handshake with the door lock within 40 cm, and the third range detects the signal and wakes up the camera; Since the camera does not match the white list face, the dynamic connection module immediately skips the first range and directly adjusts the communication power and scanning frequency to the second range, so that the door lock listens to the broadcast field and RSSI track of the courier's mobile phone within a radius of 5 m at a high frequency, which is used to generate a stranger's soft fingerprint and record the delivery time. Conversely, on the same evening, the homeowner returns home after work, and the NFC card in the key package triggers the third range when it is close to the door lock; the camera wakes up quickly and recognizes the white list face; the system immediately raises the communication parameters to the first range to complete the unlocking operation, the camera returns to sleep, and the communication power also returns to low, thereby maintaining a non-intrusive and fast unlocking experience while minimizing energy consumption.

[0060] In any of the above embodiments, the dynamic connection module comprises: a range decision unit configured to select the first range or the second range according to the result determined by the image recognition module.

[0061] a communication parameter adjustment unit connected to the range decision unit and configured to set power parameters and broadcast periods of the communication connection in the first range, the second range and the third range respectively.

[0062] In this embodiment, the range decision unit arranged in the dynamic connection module receives the identity determination result output by the image recognition module in real time. When the result is only white-listed personnel, the downstream is instructed to select the first range covering about 1 m. When the result is the presence of non-white-listed personnel or the face detection has not been started but the third range has captured an unknown device, the second range covering 5-8 m is selected. Subsequently, the communication parameter adjustment unit dynamically writes the communication stack register according to the selected range: in the first range, the transmission power of Bluetooth or Wi-Fi is set to -20 dBm and the broadcast period is 1 s; in the second range, the transmission power is increased to +4 dBm and the broadcast period is shortened to 100 ms; when the camera is in sleep mode and only the third range (30-50 cm) close to the monitoring is retained, the power is reduced to -30 dBm and the broadcast period is extended to 3 s. The adjustment unit also synchronizes the current power and period parameters to the device signal acquisition module to ensure that the scanning window and the broadcast rhythm are strictly matched, achieving the balance between optimal power consumption and maximum risk perception.

[0063] For example, the camera recognizes the face as a white list, and the range decision unit immediately selects the first range. The communication parameter adjustment unit sets the BLE power to -20 dBm and the advertising interval to 1 s, and the whole process only consumes 2 mAh to complete the unlocking. At 3 pm, a stranger salesman approaches the door, and his mobile phone is monitored in the third range. After the camera wakes up, it is determined to be a non-white list, and the range decision unit switches to the second range immediately. The adjustment unit instantaneously increases the power to +4 dBm and reduces the advertising interval to 100 ms. The device signal acquisition module acquires multiple frames of broadcast fields of the mobile phone within a radius of 8 m and generates a soft fingerprint, while triggering an alarm push to the resident's mobile phone, achieving high-sensitivity monitoring and evidence collection of suspicious visitors.

[0064] Optionally, the face detection unit can also be started automatically according to infrared detection.

[0065] In any of the above embodiments, the device signal acquisition module comprises: a signal scanning unit configured to continuously scan the signals broadcast by the nearby devices within the communication connection range and acquire the multi-dimensional information of the devices.

[0066] a soft fingerprint generation unit connected to the signal scanning unit and configured to construct and store the soft fingerprint of the device according to the multi-dimensional information containing the identity information of the device.

[0067] In this embodiment, the signal scanning unit in the device signal acquisition module continuously listens to the short-range wireless communication signals broadcast by nearby mobile terminals, including Bluetooth, Wi-Fi Probe, UWB pulse, or NFC tap-to-wake signals, within the communication connection range set by the dynamic connection module, whether it is the first range, the second range, or the low-power third range, in a multi-protocol parallel mode. Real-time analysis of signal types, bearing frequency bands, broadcast field content, RSSI time-varying curves, broadcast periods and occurrence frequencies, signal source angle of arrival (AoA) or time difference of arrival (TDoA), and other multi-dimensional information. The above raw features are then sent to the soft fingerprint generation unit: the unit first standardizes and hash encodes each dimension feature, and then combines it into a fixed-length vector V_device according to the preset weight; if the signal contains a device-side active response binding identifier (such as App authentication token or UWB Session-ID), it is written into the vector as a strong identity field; the final generated soft fingerprint is also attached with a timestamp, a door lock ID and a range marker, stored in the local ring buffer and synchronized to the cloud regularly, used as the basis data for subsequent person-device binding, repeated occurrence retrieval and abnormal trajectory analysis.

[0068] Exemplarily, when the system is in the whitelist low-power mode, the resident's mobile phone sends a BLE broadcast every second; the signal scanning unit captures the broadcast packet with a power of -20dBm, analyzes the Apple company vendor field 0x004C, the local random MAC, TX-Power, and the stable RSSI ≈-55dBm, and records the broadcast period 960ms; the soft fingerprint generation unit outputs the fingerprint vector V_A according to this, marked as a trusted device. Subsequently at 5pm, a strange Android mobile phone lingers in front of the door, its Wi-Fi ProbeRequest appears every 100ms and the RSSI rises sharply from -70dBm to -48dBm; the signal scanning unit synchronously captures its BLE extended advertisement, analyzes the vendor field 0x00E0, the random MAC change period 15min, the RSSI fluctuation feature σ=6dB, and the moving direction angle -22°. The soft fingerprint generation unit integrates multi-protocol features, generates V_B and labels it as an unknown device; if the device is recorded by different door locks again within three days, the system can automatically determine it as a suspicious repeat visitor according to the high similarity of V_B, push a risk prompt to the resident, and provide a one-key view of its previous appearance time, location, and associated face screenshot, realizing accurate tracking and evidence collection.

[0069] In any of the above embodiments, the device signal acquisition module further comprises: An identity injection interaction unit for actively broadcasting an identity handshake request signal and monitoring whether surrounding devices send a response.

[0070] The identity authentication response unit is connected with the identity injection interaction unit and is configured to receive the binding identity information sent by the response device, and send the binding identity information as the device identity information to the soft fingerprint generation unit.

[0071] In this embodiment, in addition to passive scanning, the device signal acquisition module also integrates the identity injection interaction unit and the identity authentication response unit to jointly build an active identification link of device active handshake-binding identity injection-trustworthy identity reinforcement. Specifically, the identity injection interaction unit takes the door lock as the center node, periodically broadcasts a handshake request packet through BLE GATT, a custom Wi-Fi Probe, or an optional UWBBlink frame, for example, every 3 seconds. The packet carries the door lock unique ID, a one-time random number, and a timestamp. If the surrounding mobile terminal is installed with a matching App or is authorized to implant an SDK, after receiving the handshake request, it will automatically generate binding identity information with a dynamic signature and return it through the same protocol, such as a device unique Token, a user account hash, or a secure chip signature. The identity authentication response unit is responsible for listening to and analyzing the above return data, verifying the signature validity, random number consistency, and time window legality, and then adding the extracted binding identity field to the soft fingerprint generation process of strong identity feature injection, so that the original soft fingerprint vector based on the broadcast field and RSSI is reinforced at the encryption level. Thus, without relying on the disclosure of system layer hardware identifiers, the trustworthy identification of self-App users or authorized visitor devices is realized, and the risk of uniqueness loss caused by MAC randomization is reduced.

[0072] For example, a new mobile phone is registered through the home door lock App; thereafter, the mobile phone runs a background service. When the mobile phone approaches the door lock again and receives the BLE handshake request sent by the identity injection interaction unit, the App immediately reads the random number-timeNonce in the request, performs SHA-256 signature on the device private key in the local security module, encapsulates it into the binding identity Token_0x9A17…, and returns it within 200 ms. After the identity authentication response unit verifies the signature successfully, it splices Token_0x9A17… with the captured broadcast field, RSSI curve, and other features to generate the enhanced soft fingerprint V_owner, and writes the high-trust whitelist device label for the mobile phone in the database. Two hours later, a delivery rider visits, and his mobile phone does not install the matching App and does not respond to the handshake request; the system can still extract the soft fingerprint V_rider of the rider through passive scanning, but it is marked as an unknown device due to the lack of binding identity field. If the rider delivers food multiple times and obtains temporary visitor authorization, he can be guided to install a lightweight applet to complete one-time binding, so that he can be quickly identified by Token and recorded as a low-risk visitor device during subsequent visits.

[0073] In any of the above embodiments, the identity binding module comprises: The space-time synchronization unit is configured to perform space-time matching on the face feature and the device soft fingerprint, and determine the relevance of the face feature and the device soft fingerprint.

[0074] The binding confirmation unit is connected with the space-time synchronization unit, and is configured to confirm the binding relationship based on the relevance determination result, and form binding information of the face feature and the device soft fingerprint into a binding record.

[0075] In this embodiment, the identity binding module performs double comparison on the face feature vector output by the image recognition module and the soft fingerprint vector generated by the device signal acquisition module on the millisecond time axis and the sub-meter spatial resolution through the pipeline of the space-time synchronization unit→the binding confirmation unit. The space-time synchronization unit first establishes a Δtwindow (typically 2 seconds) for each face and queries the soft fingerprint set within the window, and then calculates the time consistency degree Ct=exp(-|Δt| / τ) and the spatial consistency degree Cs=exp(-|θf–θd|² / σ²) using the face center angle θ_f provided by the door lock camera and the angle θd calculated from the angle of arrival or the RSSI gradient in the soft fingerprint, and obtains the comprehensive score S=α·Ct+β·Cs+γ·Wh by superimposing the soft fingerprint historical confidence weight Wh. The binding confirmation unit selects the soft fingerprint vector with the highest comprehensive score and S≥threshold value S_thresh for each face as the matching object, and once the binding relationship is confirmed to be established, writes the binding record containing the face vector ID, the soft fingerprint ID, the timestamp, the spatial coordinates and the comprehensive score into the binding record, and synchronizes to the local ring buffer and the cloud log chain; if all soft fingerprint scores are lower than the threshold value, an unbound flag is output, which provides a potential alarm basis for the subsequent risk module.

[0076] For example, the door lock camera captures the face of a stranger visitor and generates a vector F_x; the space-time synchronization unit opens a 2-second matching window, and receives a device soft fingerprint V_p (RSSI rises, angle -5°) after 0.3 seconds. The calculation gives Δt=0.3s, θf=-7°, θd=-5°, thus Ct≈0.86, Cs≈0.94, and since Vp is captured for the first time, Wh=0.5, and the comprehensive score S=0.86 is obtained by weighting α=0.4, β=0.4, γ=0.2. The system sets the threshold value S_thresh=0.80, and the binding confirmation unit determines The mapping is valid, and the binding entry of the stranger visitor is recorded the device SoftID #9A17… Three days later, the same soft fingerprint appears again in front of another resident's door and the camera captures a similar face, the space-time synchronization unit increases Wh to 0.9 through historical relevance self-learning, and the comprehensive score rises to 0.92, and the system automatically associates the soft fingerprint with the label of the stranger who has repeatedly appeared and pushes it to the community security platform, so that the manager can pay attention or intervene subsequently.

[0077] Specifically, the binding step of the binding confirmation unit comprises: Step one, for each face and each device signal, calculate five differences: one is the time difference between the face appearance time and the device first detection time; two is the azimuth difference between the camera estimated face direction angle and the wireless signal arrival angle; three is the dynamic difference between the face frame size rate of change over time and the signal strength rate of change over time; four is the cosine difference between the aligned face feature vector and the device fingerprint vector; five is the divergence between the respective historical appearance probability distribution of the face and the device, the execution formula includes:

[0078] In the formula, is the time difference between the face appearance time and the device signal first detection time . is the standard deviation of the time difference in the recent sliding window; is the horizontal orientation angle of the camera estimated face and the signal arrival angle measured by multiple antennas. is the historical standard deviation of the azimuth difference; is the face frame height change rate sequence within one second, used to represent the face approaching-removing speed; is the RSSI intensity change rate sequence within the same time window, used to represent the device approaching-removing speed; is the coefficient for scaling the RSSI change rate to the same dimension as the pixel change; is the dynamic time warping distance of the above two time sequences, measuring the motion synchronization degree; is the normalized face embedding vector; is the normalized device fingerprint vector; is the cross-modal projection matrix obtained by adversarial or maximum mean difference training, which maps the device fingerprint to the face embedding space; is the complement of the face-device vector cosine similarity, the smaller the value, the more similar; is the appearance probability distribution of the face in the past period of time; is the appearance probability distribution of the device fingerprint in the past period of time; is the Jensen-Shannon divergence, measuring the similarity of the two appearance modes; is the final five-dimensional difference vector, used for subsequent Mahalanobis distance and optimal transport global matching.

[0079] Step two, exponential moving average of the difference vector obtained by continuous observation, online estimation of covariance matrix, and then use Mahalanobis distance formula to map the difference vector of the last step to a single energy value. The inverse matrix of the covariance matrix can automatically adjust the weight according to the actual dispersion degree of each dimension. The rare but high-discriminative features automatically occupy a larger proportion, and the common or high-noise features automatically occupy a smaller proportion, so as to realize completely data-driven feature fusion. The execution formula includes:

[0080] In the formula, is the five-dimensional difference vector of the ith face feature and the jth device soft fingerprint, and the elements are normalized time difference, normalized orientation difference, dynamic time warping distance, face-device vector cosine difference supplement value, and jensen-shannon divergence of the historical appearance distribution of the two; is the covariance matrix obtained by exponential moving average of all difference vectors in the sliding time window; is the transpose operation of matrix or vector; is the comprehensive energy value.

[0081] Step three, arrange the energy values between all faces and all devices into a cost matrix, and send it into the entropy-regularized optimal transport model. The execution formula includes:

[0082] In the formula, is the transmission matrix, and the element represents the probability of binding face i and device j; is the cost matrix, and the element is the comprehensive energy value of the corresponding face and device. The smaller the value is, the more matched it is; is the inner product of the element-wise multiplication of the two matrices and the sum, which is used to represent the total matching cost; is the entropy regularization coefficient; is the negative entropy term of ; is the quality relaxation (penalty) coefficient, which is used to control the relaxation degree of row and column constraints; is a column vector of all 1s; is a row sum vector, which is used to represent the total probability of each face being assigned to each device; is the quality vector on the face side; is a column sum vector, which is used to represent the total probability of each device being assigned to each face; is the quality vector on the device side; is a norm, which is the sum of the absolute values of the vector.

[0083] Step four, take the maximum value of the matching probability of each face with the device, and then count the 95th percentile of all maximum probabilities on the same day as the adaptive threshold. If the maximum probability of a face is higher than the threshold, it is considered to be successfully bound; otherwise, it is marked as low confidence and triggers a risk alert, and the formula includes:

[0084] In the formula, is the optimal matching probability of face i and device j obtained by the entropy regular optimal transport algorithm; is the maximum value of the matching probability of all devices j for a fixed face i, representing the highest device probability associated with the face; is the highest matching confidence of the i-th face, and the larger the value, the more confident the system is about the highest matching confidence set of the person; is the 95th percentile of the above set.

[0085] In any of the above embodiments, the identity binding module further comprises: A risk marking unit connected with the binding confirmation unit, for marking the corresponding personnel as a potential risk personnel and generating an alarm information when the binding confirmation unit determines that the binding confidence of the face feature and the device soft fingerprint is lower than the preset threshold.

[0086] A binding database unit connected with the binding confirmation unit, for storing and managing the binding records.

[0087] In this embodiment, when the face-device matching confidence S output by the binding confirmation unit is lower than the system preset threshold Sth, the risk marking unit immediately intervenes: on the one hand, the face feature vector IDf and the personnel / device item corresponding to the soft fingerprint vector IDd are given a potential risk label, and an alarm data package containing the event timestamp, the matching score, the spatial coordinates and the on-site thumbnail is generated; on the other hand, according to the risk level, the upper alarm channel is called to push the information to the resident mobile phone App, the home control screen or the property security platform in real time, and at the same time, the dynamic connection module is triggered to enter the second range for continuous high-frequency listening, so as to capture more associated signals. Subsequently, the binding database unit writes all the binding records from the binding confirmation unit, whether successful or failed, according to the three-key index of "FaceID-DeviceID-Time" into the local ring buffer and the cloud log chain: the items with confidence ≥ Sth are marked as trusted binding, and the items below Sth are stored as risk binding, and the risk occurrence frequency of the same FaceID or DeviceID is accumulated and counted; the database implements AES-GCM encryption and hash chain verification on all records to ensure that the log is tamper-proof, and provides API for subsequent behavior clustering, blacklist updating and community big data analysis calling.

[0088] Exemplarily, a stranger wearing a cap stops at the door, the camera captures the face Fz, and the soft fingerprint is collected to the device Vm; since the face is blocked by the brim of the cap and the light is dark, the comprehensive score is only 0.52, and the binding confirmation unit determines that the pairing confidence is insufficient. The risk marking unit generates an alarm potential risk person #2023-07-01-1915 accordingly and immediately pushes it to the homeowner's mobile phone, while the Bluetooth power of the door lock is increased to +4dBm and the advertising interval is reduced to 100ms for continued listening; the binding database unit writes the record to the “risk binding” partition. Two days later, the same soft fingerprint Vm appears again in front of the neighbor's door and the matching score is still low, the system finds that the risk count of Vm reaches the threshold of 3 times, and automatically upgrades its label to high-risk device, and sends a linkage instruction to the property security background, prompting to arrange a patrol. On the contrary, a week later, the delivery man's mobile phone Vk is marked as a risk once, but then it is successfully bound with its face multiple times, the database clears the risk count of Vk and adjusts its level to a trusted visitor, avoiding repeated false positives, and fully embodies the security management benefits of the risk marking unit and the binding database unit in collaborative updating and dynamic evolution.

[0089] Embodiments of the second aspect of the application propose a method, as shown in Figure 7 comprising the following steps: S101, collecting the face image of a person in front of the door lock, extracting the face feature and comparing it with the white list; and adjusting the communication connection range to the first range or the second range according to the comparison result.

[0090] S102, scanning the wireless communication signals sent by the surrounding devices within the adjusted communication connection range, extracting multi-dimensional information and generating a device soft fingerprint; S103, spatially matching the device soft fingerprint with the corresponding face feature within the same time window to obtain the binding relationship between the face feature and the device soft fingerprint, and storing it as a basis for personnel identity recognition.

[0091] The method provided by the application determines the communication connection range by comparing with the face white list (S101), generates a device soft fingerprint within the adaptive range, and completes face-device spatial matching and database storage within the same time window. The method realizes power saving, fast release and high risk perception in a closed loop: when familiar people are detected, small radius and low power communication is maintained to ensure non-intrusive unlocking and prolong battery life; once strangers appear, the communication range is instantaneously enlarged, enabling the capture of signals such as mobile phones / Wi-Fi / UWB at a distance and the construction of a soft fingerprint, which is then bound with the image synchronously to form a traceable person-device double feature file. Thus, the defects of low-resolution cameras in independently marking strangers are remedied, and the waste of battery life caused by fixed high-power listening is avoided, thereby improving the real-time recognition accuracy, risk alarm timeliness and post-evidence reliability of home security.

[0092] The integrated modules / units, if implemented in the form of software functional units and sold or used as independent products, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by instructing related hardware through a computer program, and the computer program can be stored in a computer readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned various method embodiments can be implemented. The computer program can include computer program code, which can be in the form of source code, object code, executable file or some intermediate form. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), electric carrier wave signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electric carrier wave signal and telecommunication signal.

[0093] The above embodiments are only used to illustrate the technical solutions of the present disclosure, rather than limit them; although the present disclosure has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should be included in the protection scope of the present disclosure.

Claims

1. An intelligent home security system based on the Internet of Things, characterized in that: include: An image recognition module is used to collect facial images of people in front of the door lock and extract facial features, compare the facial features with a pre-stored whitelist, and determine whether the person is on the whitelist; a dynamic connection module connected to the image recognition module, configured to set the communication connection range to a first range when the image recognition module detects that only whitelisted persons are present in front of the door lock, or to expand the communication connection range to a second range when the image recognition module detects that non-whitelisted persons are present; a device signal acquisition module, connected to the dynamic connection module, configured to scan and acquire wireless identification signals of surrounding devices within the communication connection range, extract multi-dimensional information of the devices, and construct a soft fingerprint of the devices; The identity binding module is connected to the image recognition module and the device signal acquisition module respectively, and is used to perform spatiotemporal synchronization matching between the facial features and the device soft fingerprint, generate a binding relationship between the facial features and the device soft fingerprint, and store the binding relationship for personnel identity recognition.

2. The smart home security system according to claim 1, characterized in that: The image recognition module includes: A face detection unit is used to detect faces in the area in front of the door lock in real time when in the wake-up state, and determine the location and number of faces; a feature extraction unit, connected to the face detection unit, for performing feature embedding on the detected face and extracting a face feature vector; The identity comparison unit is connected to the feature extraction unit and is used to calculate the similarity between the facial feature vector and the pre-stored white list to determine whether the person corresponding to the face is a white list person.

3. The smart home security system according to claim 2, characterized in that: The communication connection range of the dynamic connection module also includes a third range. When a device appears in the third range, the face detection unit is started and is in an awake state.

4. The smart home security system according to claim 3, characterized in that: The spatial size of the third range is smaller than that of the first range, and when the face detection unit is in an awake state, the communication connection range is changed from the third range to the first range or the second range.

5. The smart home security system according to claim 3, characterized in that: The dynamic connection module includes: a range decision unit, configured to select the first range or the second range according to a result determined by the image recognition module; A communication parameter adjustment unit is connected to the range decision unit and is used to set the power parameters and broadcast period of the communication connection in the first range, the second range and the third range respectively.

6. The smart home security system according to claim 1, characterized in that: The device signal acquisition module includes: A signal scanning unit is used to continuously scan the signals broadcast by nearby devices within the communication connection range and collect multi-dimensional information about the devices; A soft fingerprint generating unit is connected to the signal scanning unit and is used to construct and store the device soft fingerprint according to the multi-dimensional information including the device identity information.

7. The smart home security system according to claim 6, characterized in that: The device signal acquisition module also includes: The identity injection interaction unit is used to actively broadcast the identity handshake request signal and monitor whether the surrounding devices respond; The identity authentication response unit is connected to the identity injection interaction unit, and is used to receive the binding identification information sent by the response device, and send the binding identification information as the device identity information to the soft fingerprint generation unit.

8. The smart home security system according to claim 1, characterized in that: The identity binding module includes: a spatiotemporal synchronization unit, configured to perform spatiotemporal matching between the facial features and the device soft fingerprint, and determine the correlation between the facial features and the device soft fingerprint; The binding confirmation unit is connected to the spatiotemporal synchronization unit and is used to confirm the binding relationship based on the association determination result, and form a binding record with the binding information of the facial features and the device soft fingerprint.

9. The smart home security system according to claim 8, characterized in that: The identity binding module also includes: a risk marking unit, connected to the binding confirmation unit, configured to mark the corresponding person as a potential risk person and generate an alarm message when the binding confirmation unit determines that the binding confidence level between the facial features and the device soft fingerprint is lower than a preset threshold; The binding database unit is connected to the binding confirmation unit and is used to store and manage the binding records.

10. A method implemented by the smart home security system according to any one of claims 1 to 9, characterized in that: The steps include: Collecting a facial image of a person in front of the door lock, extracting facial features and comparing them with the whitelist; adjusting the communication connection range to the first range or the second range according to the comparison result; Scanning wireless communication signals sent by surrounding devices within the adjusted communication connection range, extracting the multi-dimensional information and generating the device soft fingerprint; The device soft fingerprint is spatially matched with the corresponding facial features within the same time window to obtain a binding relationship between the facial features and the device soft fingerprint, and is stored as a basis for personnel identification.

Images