Vibration sensor and electronic control comprising such a sensor

Abstract

A vibration sensor for a door or window is provided and configured to detect an intrusion, the vibration sensor comprising a processor configured to apply a machine learning model to process the sensor data so as to determine the presence of anomalies in the data and, in the event an anomaly is determined, to generate an output.An electronically controlled lock comprising such a vibration sensor is also provided, the sensor comprising at least one accelerometer or magnetometer, wherein the processor is configured to execute a first classifier to detect anomalies in event-related signals received from the accelerometer or magnetometer, and upon detecting an anomaly, to execute a second classifier to attempt to classify the event into one of multiple event classes, at least one of the classes indicating an alarm event and at least one of the classes corresponding to a non-alarm event, the processor being configured to signal an alarm event in the event that the classifier determines that an event is an alarm event, and optionally also in the event that an event cannot be classified as a non-alarm event. Figure 4A.

Description

Title of the invention: Vibration sensor and electronic control comprising such a sensor Technical field

[0001] The present invention relates to a vibration sensor, an electronically controlled lock for a premises door, the lock comprising such a sensor, premises security monitoring installations comprising one or more such sensors, and / or locks, as well as associated systems, subsystems and methods.

[0002] The electronically controlled lock may be in the form of, or include, an electronic lock driver with an electric motor or solenoid to drive the mechanism of a mechanical lock, rather than being a fully integrated smart lock. Context

[0003] Locks intended to protect the doors of premises to prevent their illicit opening and access to the interior of the premises (access control), have been known for hundreds, if not thousands of years, the general idea being to prevent anyone without a suitable key from accessing the protected premises secured by the door. But for as long as there have been locks, there have always been criminals who try to open them without the appropriate key, for example by using lock picks or a "strike key". In the last decade, electromechanical locks, called smart locks, have been introduced with the aim of improving security and possibly user convenience.Such smart locks typically take the form of an electronic token or electromagnetically signaled control, replacing the usual key-lock mechanism, which can control the locking or unlocking of a locking mechanism. Such locks may operate with access cards or other tokens, or with active devices such as phones, which can generate RF signals or be read by RF signals (e.g., Bluetooth, NFC, RFID, or others), or magnetically, and / or a biometric interface (e.g., an iris recognition, fingerprint reading, or facial recognition interface). Many smart lock designs lack a keyhole, so that activation of the lock can only be implemented using electromagnetic means.Many home users seem reassured by the presence of a permanent keyhole that cooperates with a tangible key that can be carried by the user.

[0004] Smart locks can be used as independent security measures, and such a lock may include a buzzer that is activated in the event that the lock detects an attempt at manipulation. But smart locks can also be used in or in conjunction with premises security monitoring systems, as well as in conjunction with home automation installations, often with the ability to report any detected attempt at manipulation or interference. Such a smart lock, which may also be referred to as an electronically controlled lock, is described in the applicant's earlier patent application first published as WO2023 / 227258.This smart lock is designed to meet the requirements of EN50131 Grade 2, so the smart lock can be integrated into a security monitoring system that otherwise meets the requirements of EN50131 Grade 2 or higher.

[0005] Some smart locks are designed to work with conventional lock cylinders, providing either a convenient upgrade to an existing lock installation or a way to enjoy both the convenience of electronic lock control and the ability to continue using mechanical keys. These smart locks, however, may be as susceptible to "knock-in" attacks as the mechanical lock cylinders on which they are based.

[0006] In this patent application, the terms "smart lock" and "electronically controlled lock" should be interpreted to encompass conventional (mechanical) locks to which an electronic lock driver has been added. Such electronic lock drivers may include an electric motor or solenoid that is mechanically coupled to a mechanical lock such that the electronics of the lock driver can be used to control the operation of (or "manipulate") the lock. The terms "smart lock" and "electronically controlled lock" also encompass an electronic lock driver for mounting on a mechanical lock assembly.

[0007] Although there are many designs of smart locks, there is a need for an improved smart lock, particularly one suitable for installation in domestic premises. It is also desirable that such a smart lock meets the requirements of EN50131 Grade 2, such that the smart lock can be integrated into a security monitoring system that otherwise meets the requirements of EN50131 Grade 2 or higher. Preview

[0008] Various aspects of the invention are described herein. These aspects may be used independently, or two or more aspects may be used in combination.

[0009] A first aspect relates to an electronically controlled lock for a premises door, the lock comprising a processor and an accelerometer or magnetometer coupled to the processor, wherein: the accelerometer or magnetometer is configured to operate as a vibration sensor, the processor and the accelerometer / magnetometer operating together as a vibration detector to generate an alert signal in the event of an attempt to pick or interfere with the lock;and the processor is configured to first execute a first classifier to detect anomalies in the signals received from the accelerometer / magnetometer, during or following an event, and in the event of detection of an anomaly, to execute a second classifier to classify the anomaly as belonging to an identified event class among a plurality of event classes, and optionally wherein the processor is configured to transmit an alert signal using a transceiver of the lock, in the event of an attempt to pick or interfere with the lock. ;

[0010] A second aspect relates to a smart lock comprising: an accelerometer or magnetometer configured as a shock detector, a processor configured to detect an anomaly in a signal received from the accelerometer or magnetometer, the processor being configured to implement an autoencoder, the autoencoder being configured to classify the received signal based on pre-trained non-anomalous signal data and to re-estimate the corresponding signal, and the processor being configured to detect an anomaly based on a difference between the received signal and the re-estimated signal. Optionally, the processor may be configured (e.g. programmed) to detect an anomaly based on an energy difference between the received signal and the re-estimated signal.

[0011] A third aspect relates to an electronically controlled smart lock for a premises door, the lock comprising an accelerometer or a magnetometer, a processor coupled to the accelerometer / magnetometer and configured to execute a first classifier to detect anomalies in signals received from the accelerometer / magnetometer with respect to an event, and upon detecting an anomaly, to execute a second classifier to attempt to classify the event into one of multiple event classes, at least one of the classes signifying an alarm event and at least one of the classes corresponding to a non-alarm event, the processor being configured to signal an alarm event in the event that the classifier determines that an event is an alarm event, and optionally in the event that an event cannot be classified as a non-alarm event.

[0012] A fourth aspect of the invention relates to a vibration sensor (e.g. for a door or window and configured to detect an intrusion), the vibration sensor vibration comprising a processor that is configured to apply a machine learning model to process the sensor data so as to determine the presence of anomalies in the data, and in the event that an anomaly is detected / determined, to generate an output. The vibration sensor may comprise an accelerometer and / or a magnetometer, and optionally a microphone. The microphone may be placed inside the lock and arranged to detect sounds of attacks on the lock, such as lock picking. A further microphone may be provided in the lock or in a second detection arrangement to detect ambient sounds, and in particular ambient sounds on the exposed or at-risk side of the door. These sounds may be used to help distinguish false alarms from real alarm events (e.g.thanks to the signals representing such sounds, which are made available to an entity monitoring the installation of which the lock and / or the second detection arrangement are part and which are processed by this entity).

[0013] Any microphone forming part of the lock or the second sensing arrangement is preferably placed in the housing of the lock or the second sensing arrangement in such a way that the microphone is protected against direct attack through any opening, port or grille through which ambient sound reaches the microphone - for example by offsetting the microphone from such a port, opening or grille, for example by being mounted transversely thereto, so that the microphone is not exposed to damage in the event that a foreign object (such as a screwdriver or drill) is inserted through the opening, port or grille.A microphone for detecting the sound of lock picking may be coupled through an open passage to a keyhole of the lock, but is preferably again offset, e.g. mounted transversely to the axis of the keyhole bore, so that the microphone cannot be easily attacked through the keyhole.

[0014] In some embodiments, the processor is programmed to operate as an autoencoder to process the sensor data.

[0015] In some embodiments, for example, the autoencoder is configured to adjust and / or classify a received signal based on pre-trained non-anomalous signal data and to re-estimate the corresponding signal, and the processor is configured to detect an anomaly based on a difference between the received signal and the re-estimated signal. The processor may be configured to detect an anomaly based on an energy difference between the received signal and the re-estimated signal.

[0016] In some embodiments, in the event that the processor determines the existence of an anomaly, the processor is programmed to analyze the sensor data, or the processed sensor data, to attribute a detected event to one of the multiple event types, the event types including at least one threat class and at least one non-threat class. The threat class may include at least one of lock picking, lock tapping, and lock picking.

[0017] A fifth aspect of the invention relates to an electronically controlled lock comprising a vibration sensor as described above, and optionally wherein the smart lock comprises an intelligent electronic lock driver operatively coupled to a mechanical lock.

[0018] The lock may include a keyhole and the machine learning model may be trained to recognize as abnormal events at least one of lock picking, lock tapping and lock picking. A lock having a keyhole may include a microphone, for example arranged to preferably capture sound from a keyhole of the lock and / or from the vicinity of a lock cylinder, and the digital processor circuit may be arranged to receive signals or data from the microphone in addition to signals received from the accelerometer or magnetometer;optionally, the lock may further or alternatively comprise a pressure sensor and / or an insertion detector (possibly based on the mechanical displacement of a sensing element) for detecting the insertion of an element (such as a key, a strike key or a pick) into an external keyhole of the lock, and the digital processor circuit (or a remote processor) may be arranged to receive signals or data from one or more of these sensors, in addition to the signals received from the accelerometer or the magnetometer or simply as additional sensors whose signals may contribute to the determination of an abnormal event and possibly to its classification. ;

[0019] The lock may further comprise one or more electronic sensing arrangements for implementing hand-on-handle or hand-on-lock detection, e.g., capacitive sensing, and / or key-in-lock detection (e.g., using impedance sensing), and the digital processor circuit is configured to use data from the one or more electronic sensing arrangements to classify events into particular attack modes, such as lock picking, lock picking, and lock kicking.

[0020] A sixth aspect of the invention relates to an electronically controlled lock comprising a vibration sensor as described above, wherein the processor is configured to implement an auto-encoder, the auto-encoder being configured to classify a signal received from the accelerometer or magnetometer based on pre-trained non-anomalous signal data and to re-estimate the signal corresponding, and the processor is further configured to detect an anomaly based on a difference between the received signal and the re-estimated signal. The processor may be configured to detect an anomaly based on an energy difference between the received signal and the re-estimated signal.

[0021] A seventh aspect relates to an electronically controlled lock for a premises door, the lock comprising a processor and, coupled to the processor, an accelerometer or magnetometer, wherein: the accelerometer or magnetometer is configured to operate as a vibration sensor, the processor and the accelerometer or magnetometer operating together as a vibration detector to generate an alert signal in the event of an attempt to pick or interfere with the lock;the processor is first configured to detect anomalies in the signals received from the accelerometer or magnetometer, during or following an event, and upon detection of an anomaly, the processor is configured to execute a classifier to classify the anomaly as belonging to an identified class among a plurality of event classes, and optionally wherein the processor is configured to transmit an alert signal using a transceiver of the lock, in the event of an attempt to pick or interfere with the lock. ;

[0022] The locks according to the seventh aspect may comprise a keyhole and may further comprise one or more additional sensors in the form of:

[0023] (i) a microphone, for example arranged to preferentially capture sound from a keyhole of the lock and / or from the vicinity of a lock cylinder, the digital processor circuit being arranged to receive signals or data from the microphone in addition to signals received from the accelerometer or magnetometer;

[0024] (ii) a pressure sensor; and / or

[0025] (ii) an insertion detector (optionally based on mechanical displacement of a sensing element) for detecting insertion of an element (such as a key, a strike key or a pick) into an external keyhole of the lock:

[0026] and the processor (or a remote processor) may be arranged to receive signals or data from any such additional sensor in addition to the signals received from the accelerometer or magnetometer, the digital processor circuit (or a remote processor) being programmed or configured to use signals or data from any such additional sensor for the determination and / or classification of an abnormal event.

[0027] An eighth aspect of the invention relates to a premises security monitoring installation comprising a control unit for controlling the operation of the security monitoring installation, the installation comprising: a window or door vibration sensor incorporating a first detector, the first detector comprising an accelerometer or a magnetometer; an electronically controlled lock according to any variant of the seventh aspect incorporating a second detector, the second detector comprising an accelerometer or a magnetometer; the window or door vibration sensor and the smart lock being configured to transmit signal data from the first and second detectors to the control unit; and the control unit being configured / programmed to classify an event based on the signal data received from the first and second detectors, and optionally based on signal data from one or more of the aforementioned additional sensors of the lock.

[0028] A ninth aspect of the invention relates to a premises security monitoring installation comprising a control unit for controlling the operation of the security monitoring installation, the installation comprising: an electronically controlled lock according to any variant of the seventh aspect coupled to a door of the premises, the electronically controlled lock incorporating a first detector, the first detector comprising an accelerometer or a magnetometer; a door vibration sensor having incorporated therein a second detector, the second detector comprising an accelerometer or a magnetometer, coupled to the same door at a position remote from the lock; the door vibration sensor and the smart lock being configured to transmit signal data from the first and second detectors to the control unit; and

[0029] the control unit being configured / programmed to classify an event based on signal data received from the first and second detectors, and optionally based on signal data from one or more of the aforementioned additional sensors of the lock.

[0030] A tenth aspect of the invention relates to an installation comprising an electronically controlled lock mounted on a door, the electronically controlled lock being in accordance with any variant of the seventh aspect and comprising a first detection arrangement in the form of a first accelerometer and / or magnetometer, the installation comprising a unit, separate from the lock, mounted on the door or on a door frame which receives the door, the unit comprising a second detection arrangement separate from the first detection arrangement, the second detection arrangement comprising one or more of an accelerometer and a magnetometer, the installation comprising a digital processor circuit configured to receive signals from the first and second detection arrangements, and optionally from one or more of the aforementioned additional sensors of the lock,and to process the signals to detect a potential intrusion attempt based on data detected simultaneously from different detection positions relative to the door and / or door frame.

[0031] An eleventh aspect of the invention relates to an electronically controlled lock comprising a vibration sensor as previously described, wherein the processor is configured to execute a first classifier to detect anomalies in the signals received from the accelerometer or magnetometer with respect to an event, and upon detection of an anomaly, to execute a second classifier to attempt to classify the event into one of multiple event classes, at least one of the classes signifying an alarm event, and at least one of the classes corresponding to a non-alarm event, the processor being configured to signal an alarm event in the event that the classifier determines that an event is an alarm event, and optionally also in the event that an event is not classifiable as a non-alarm event.

[0032] A twelfth aspect of the invention relates to a premises security monitoring installation comprising at least one vibration sensor having any of the above characteristics, and / or an electronically controlled lock as described above.

[0033] A thirteenth aspect of the invention relates to a premises security monitoring installation, optionally as above, comprising a control unit for controlling the operation of the security monitoring installation, the installation comprising: a window or door vibration sensor incorporating a first detector, the first detector comprising an accelerometer or a magnetometer; an electronically controlled lock, optionally as previously described, incorporating a second detector, the second detector comprising an accelerometer or a magnetometer; the window or door vibration sensor and the smart lock being configured to transmit signal data from the first and second detectors to the control unit; and the control unit being configured / programmed to generate a score based on the signal data received from the first and second detectors.

[0034] A fourteenth aspect of the invention relates to a premises security monitoring installation optionally as defined above, comprising a control unit for controlling the operation of the security monitoring installation, the installation comprising: an electronically controlled lock, optionally as defined above, coupled to a door of the premises, the electronically controlled lock incorporating a first detector in the form of an accelerometer or a magnetometer; a door vibration sensor, incorporating a second detector in the form of an accelerometer or a magnetometer, coupled to the same door at a position remote from the lock; the door vibration sensor and the smart lock being configured to transmit signal data from the first and second detectors to the control unit; and the control unit being configured / programmed to classify an event based on signal data received from the first and second detectors.

[0035] A fifteenth aspect of the invention relates to a premises security monitoring installation, optionally as described above, wherein the control unit comprises a processor configured to apply a machine learning model, such as an autoencoder, to process received signal data to determine the presence of anomalies in the received signal data.

[0036] A sixteenth aspect of the invention relates to an installation comprising an electronically controlled lock mounted on a door, the electronically controlled lock comprising a first detection arrangement in the form of an accelerometer and / or a magnetometer, the installation comprising a unit, separate from the lock, mounted on the door or on a door frame which receives the door, the unit comprising a second detection arrangement which comprises an accelerometer and / or a magnetometer (such that the first and second detection arrangements may each comprise the same type of detector or different types of detector in either order),the installation comprising a digital processor circuit configured to receive signals from the first and second detection arrangements and to process the signals to detect a potential intrusion attempt based on data detected simultaneously from different detection positions relative to the door and / or door frame.

[0037] The processor circuit may include a processor programmed to operate as an autoencoder to process data from the first and second sensing arrangements.

[0038] In some embodiments, in the event that the processor determines that an anomaly exists, the processor circuitry is programmed to analyze the sensor data, or the processed sensor data, to associate a detected event with one of multiple event types, the event types including at least one threat class and at least one non-threat class. The threat class may optionally include at least one of lock picking, lock tapping, and lock picking.

[0039] In some embodiments, the processor circuit includes a processor that is a component of the electronically controlled lock.

[0040] The installation may further comprise a security monitoring system for the premises of which the door is a part, and optionally at least a portion of the processor circuit may comprise a processor of a controller of the security monitoring system. The controller of the security monitoring system may optionally be provided with a video doorbell arrangement.

[0041] The security monitoring system may have at least one mode of operation in which it is arranged to notify alarm events to a remote monitoring entity. The security monitoring system may for example have at least one mode of operation in which it is arranged to notify the remote monitoring entity of abnormal events detected in the sensor data.

[0042] A seventeenth aspect of the invention relates to a premises security monitoring installation comprising a control unit for controlling operation of the security monitoring installation, the security monitoring installation further comprising: a door vibration sensor incorporating an accelerometer and / or a magnetometer, the vibration sensor being coupled to a door of the premises; an electronically controlled lock incorporating an accelerometer or a magnetometer, the electronically controlled lock being coupled to the door of the premises; the vibration sensor and the electronically controlled lock being configured to transmit signal data derived from their accelerometer(s) and / or magnetometer(s) to the control unit; and the control unit being configured / programmed to classify an event based on the signal data derived from the first and second vibration sensors.

[0043] The control unit may include a processor that is configured to apply a machine learning model, such as an autoencoder, to process the received signal data to determine the presence of anomalies in the received signal data.

[0044] The control unit may be configured / programmed to provide a quantized output whose value is selected from at least three levels (i.e., non-binary).

[0045] A further eighteenth aspect of the invention relates to a method of operating an installation comprising an electronically controlled lock mounted on a door, the electronically controlled lock comprising a first sensing arrangement in the form of an accelerometer and / or a magnetometer, the installation further comprising a unit, separate from the lock and spaced apart from the lock, mounted on the door or on a door frame which receives the door, the unit comprising a second sensing arrangement (separate from the first sensing arrangement) which comprises an accelerometer and / or a magnetometer, the installation further comprising a digital processor circuit, the method comprising: processing by the digital processing circuit signals derived from the first sensing arrangement (e.g. a first accelerometer) and the second sensing arrangement (e.g. a second accelerometer) in order to detect events detectable simultaneously from different detection positions relative to the door and / or door frame.

[0046] The method may further comprise combining output data from the first and second sensing arrangements and processing the combined output data, optionally wherein the output data from each sensing arrangement comprises accelerometer output data.

[0047] Additionally or alternatively, the method may further comprise processing the output data from the first detection arrangement by the first detection arrangement, processing the output data from the second detection arrangement by the second detection arrangement, and combining the processing results on a central processing unit of the installation.

[0048] Non-limiting features and advantages of the invention include the ability to detect abnormal vibrations, for example, relative to vibrations detectable under normal operating conditions of the smart lock and door. When a machine learning algorithm is used and trained only on training data under "normal" conditions or by prioritizing them, it is possible to detect any suspicious vibration as an anomaly, even if it corresponds to an intrusion attempt or an attack on the lock that the system has not been trained to specifically recognize (e.g., a new type of intrusion attempt or lock attack).When multiple sensing arrangements are used, one on the lock and another at a different position distant from the lock, additional context can be inferred by sensing the same vibration event substantially simultaneously from or at multiple sensing positions. This can facilitate automated discrimination between an intrusion attempt targeting the lock and a different intrusion attempt targeting another part of the door.

[0049] This may also help to reduce the occurrence of false alarms compared to using a single detection arrangement on the door.

[0050] Additional aspects of the invention are defined in the claims. Independent protection is also claimed for any novel features and / or ideas disclosed herein and / or in the drawings, whether or not emphasized. Brief description of the figures

[0051] Embodiments of the invention will now be described, by way of example only, with reference to the accompanying figures, in which:

[0052] [Fig.l] schematically shows premises equipped with a smart lock installation which may include video surveillance of the areas surrounding the smart lock.

[0053] [Fig.2] schematically illustrates the main components of a smart lock according to one aspect of the invention;

[0054] [Fig.3A] shows a mortise lock of a type with which aspects of the invention may be used, installed in a door;

[0055] [Fig.3B] shows an assembly which includes the mechanism and electronics of an electronic lock according to aspects of the invention;

[0056] [Fig.3C] schematically shows a longitudinal section through the whole of [Fig.3B];

[0057] Figure 4 schematically illustrates methods according to aspects of the invention;

[0058] [Fig.5] schematically illustrates the elements which constitute an auto-encoder as it can be used in certain aspects of the invention;

[0059] [Fig.6] schematically illustrates how an autoencoder trained according to aspects of the invention can distinguish between "routine" events and aberrant events;

[0060] [Fig.7] schematically illustrates how an auto-encoder can be integrated into a smart lock according to one aspect of the invention;

[0061] [Fig.8] schematically illustrates the main constituents of a video doorbell according to one aspect of the invention;

[0062] [Fig.9] schematically illustrates a home security monitoring system built around a video doorbell; and

[0063] Figure 10 generally corresponds to [Fig.9], but also includes the advantages of professional monitoring by means of a remote alarm receiving center. Specific description

[0064] [Fig.l] schematically shows a home (or "premises") with a smart lock installation that may include video surveillance of the area surrounding the smart lock. In this case, a video doorbell 100 may be provided adjacent to the main door 102, which here is the front door of the house 104. The front door 102 is equipped with a "smart" lock 106 that may be locked and unlocked by electrical control signals, received for example from a personal communication device 108 of the home owner. The video doorbell 100 may for example cooperate with a backend service 110 that sends push notifications to a user's personal communication device 108 when the doorbell button is pressed (or optionally when someone approaches the doorbell 100 or another video surveillance device).To support such communications, the video doorbell can use Wi-Fi to connect to a Wi-Fi router 112 (or a controller of a premises security monitoring facility) that has a connection. broadband 114 to the Internet 115, the backend service 110 sending push notifications to a Public Land Mobile Network (PLMN) 116 by means of which notifications can be delivered to the user's personal communication device 108. The same communication path can be used to stream live video (and audio) from the video doorbell 100 to the user. A reverse communication path can also be supported to allow the user to speak to a person at the doorbell 100, and also to allow the user to activate (i.e., unlock) the lock 106 to allow the person at the door to open the door 102.The signals to activate (unlock) the lock may be transmitted directly to the lock 106 from the router 112 (or the security monitoring system controller), or they may be transmitted via the video doorbell 100 which in turn communicates with the smart lock 106. The smart lock 106 has been designed to comply with grade 2 of EN50131 by incorporating a shock sensor (or more generally a vibration sensor) with a tamper detection function.

[0065] Additionally or alternatively, the smart lock, video doorbell and security monitoring system controller may additionally or alternatively be provided with at least one transceiver supporting NB-IOT (narrowband Internet of Things, supported by 3GPP), LTE-M (Long-Term Evolution Machine Type Communication) or other similar protocols, which are LPWAN (Low Power Wide Area Network) technologies that provide secure connections using licensed spectrum in carrier-managed networks.

[0066] [Fig. 2] schematically illustrates the principal components of a smart lock 106, such as that shown in [Fig. 1], according to embodiments of the invention. The lock 106 is shown installed in a door 102, with a door frame 240 receiving a cooperating strike 242 in which at least one bolt 244 is received. The lock 106 may be configured as a mortise lock, in which case the lock will typically include a mortise lock assembly or a latch (not shown) in addition to the bolt 244.

[0067] Providing the lock 106 as a mortise lock is particularly advantageous if the lock is to be pre-installed in a door (or if the door is to be manufactured with mortises provided for easy installation of the lock on site) rather than being retrofitted to a pre-hung door as a replacement for an existing lock. But there is generally a desire, particularly when replacing a surface-mounted lock, to surface-mount the lock 106 (on the protected side of the door 102, typically its inner side), because Installation is typically much easier and faster if a cavity (mortise) does not have to be created in the door to accept the new lock. In either case, the bolt 224 can be received in a cavity inside the door, but with a surface-mounted lock, the bolt can also be on the outside of the door. The lock 106 is provided with an actuator 246, which may be a solenoid or a motor, driving the bolt into the locked or unlocked position, or engaging / disengaging a clutch which couples / uncouples a knob or handle 248 on the exposed face of the door and by means of which the bolt can be pulled into the unlocked position allowing the door to be opened from the exposed (unprotected) side. Typically, if the lock 106 is provided with a clutch handle, the actuator will be a solenoid rather than a motor.The lock 106 may of course be provided with another handle or another button (not shown), on the protected side of the door 102, to actuate the bolt 244 in order to allow the opening and locking of the door 102 from the protected (generally internal) side of the door 102.

[0068] Although, for ease of illustration and explanation, [Fig. 2] illustrates a lock installation with a single bolt 244 and a corresponding single strike 242, locks according to embodiments of the invention may comprise, or be incorporated into, or used with, multi-point lock installations / doors where more than one bolt is actually used to secure a door.

[0069] The lock 106 includes a power supply 250 which preferably includes a rechargeable battery 252, and preferably also a charging arrangement 254 to facilitate on-site charging of the rechargeable battery 252. The charging arrangement may be coupled to a mains power supply, but for ease and simplicity, it may be preferable to include an interface 256 (such as a micro-USB port, a USB-C port, a "Lightning" jack, or the like) to accept a low voltage supply (e.g., 5 V or 20 V, from an external battery or charger) which, in the absence of a mains supply (or in the event of a mains supply failure), can be used to recharge the rechargeable battery 252 on-site, rather than requiring removal of the battery for recharging, such that the lock 106 can still be powered.In another embodiment, the lock may include a coil or other arrangement allowing inductive charging from a suitably placed transmitter connected to an external power source (such as a battery or mains power).

[0070] The actuator 246 is operatively coupled to and controlled by a processor 258, the processor having an associated memory 260 in which are stored program instructions ("software") which, when executed on the processor 258, control the operation of the lock 106. The processor 258 may for example be a microcontroller or a microprocessor. At least one transceiver 262, which is configured to receive control signals from for example a controller of the security monitoring system (e.g. from the video doorbell 100 or the controller of the security monitoring installation in the embodiment shown in [Fig. 1]) and to transmit event information thereto, is also coupled to and controlled by the processor 258. These may be relatively low bandwidth transmissions (e.g. in comparison with video transmissions of the video doorbell 100), so a low bandwidth channel is suitable - which means that a low power consumption transceiver can be used, thus providing a sufficiently long battery life.The transceiver 262 may, for example, be configured to operate using an appropriately allocated frequency in the Industrial, Scientific, and Medical (ISM) bands—such as 868 MHz (in Europe). Communications to and from the lock are preferably encrypted. As previously noted, the transceiver 262 may additionally or alternatively also support a Low Power Wide Area Network (LPWAN) technology, such as NB-IOT or LTE-M, for communicating with one or more user devices (WTRUs) 108 and / or a remote backend system or monitoring station, or an Alarm Receiving Centre (ARC).

[0071] The lock 106 may also include a chime or speaker 263 (more generally, a sound output device, e.g., an electromechanical chime or buzzer), which is operatively coupled to the processor 258. This speaker may be used instead of or in addition to a separate chime (e.g., located in a lobby or elsewhere on the premises) that sounds whenever a chime triggering device of the video doorbell is activated - a processor of the video doorbell 100 being optionally configured to use a transceiver to transmit a chime activation signal labeled with a marking to which the lock processor 258 and / or the processor of a separate chime ringing device (not shown) responds by using an internal sound output device to sound a chime in response to activation of the chime triggering device.

[0072] The lock 106 may be provided with a housing 259 through which the bolt 244 and any mortise lock assembly may protrude, with the knob or handle 248 of course being outside the housing 259. The housing is preferably made of an engineering plastic material or a non-ferromagnetic material so as not to interfere with the operation of one or more magnetometers optionally forming part of the lock 106.

[0073] The lock 106 may further include a magnetometer 264 that may be configured to detect and monitor a first magnetic field produced by an optional first magnet 266 that may be fixedly mounted relative to the frame 240 or other stop against which the door 102 closes. The same magnetometer 264, or an additional magnetometer (not shown), may also be configured to detect and monitor a second magnetic field produced by a second magnet 270 that may be attached to the door, and preferably mounted on the inside of the door - i.e., between the protected face and the unprotected face of the door, although it is also possible to attach (e.g., with an adhesive and / or a mechanical fastener such as a screw) a magnet 270 to the face of the door (on the inside of the door, e.g., the inside of the building).This second magnet 270, if present, is external to the lock 106 and is provided such that a magnetometer configured to detect the magnetic field of the second magnet 270 can generate a tamper signal to indicate an attempt to remove the lock 106 from the door 102. On the other hand, a magnetometer 264 can be configured to detect and monitor the first magnetic field produced by a first magnet 266 and can send a signal to indicate the state of the door, which indicates for example whether the door 102 is open or closed.The processor 258 may be configured to generate an alert signal, which is transmitted to a controller of the security monitoring system, for example an improved video doorbell according to certain aspects of the invention, for subsequent reporting to a monitoring service and / or the owner / occupant of the home, in the event that the opening of the door is detected without the lock having been unlocked and / or the alarm having been disarmed - since such a combination of circumstances may indicate that the door 102 has been forced.It is noted that a magnetometer can function as a shock or vibration sensor to the extent that it is able to respond to short-term disturbances, or disturbances in the magnetic field from an associated magnet, for example, as a result of the magnetometer moving relative to the source of the magnetic field: a magnetometer mounted on or in a door can, for example, register a shock caused by an object striking the door (e.g., a thief trying to gain entry, or a football being forcefully kicked against the door) due to small variations in the spacing between the magnetometer and the source of the magnetic field. It is therefore conceivable to use a magnetometer instead of an accelerometer when designing a vibration sensor. In general, embodiments of the present invention can be implemented using either an accelerometer or a magnetometer, or both.Depending on the application and the precise details of any installation, an accelerometer may be more sensitive to small disturbances, and therefore it may be preferable to use an accelerometer rather than a . magnetometer. But in other situations, a magnetometer may be preferred because of its relative insensitivity to minor disturbances.

[0074] The lock 106 may further include an accelerometer 275, which is also coupled to the processor 258 and which is configured to function as a shock sensor (or more generally a vibration sensor) to provide the processor with signals enabling the processor to detect attempts to "pick" or "hit" the lock, drill the lock cylinder, and other mechanical attacks on the lock, as well as to provide signals in the event of attempts to break down the door (or force the lock), with a sledgehammer or a battering ram, for example.The processor 258 preferably uses machine learning, e.g., deep learning, and / or other artificial intelligence approaches to distinguish accelerometer signals generated as a result of “everyday events,” such as a normal knock on the door such as “I’m here!”, unlocking the lock with the appropriate key, slamming an unlocked door, accidental impact of a soccer ball, etc., from anomalous signals generated as a result of malicious activity, such as an attempt to pick or kick a lock, or attempts to force or break down the door. The application of artificial intelligence techniques to this problem of discriminating between signals representing anomalous events and “everyday,” “non-abnormal,” or “ordinary” events is described and discussed later in this patent application.The 275 accelerometer (also referred to herein as the "alarm device") also provides a potential basis for certifying the lock as a shock detector (for at least a grade 2 or 3 alarm system) according to EN 50131-2-8. The accelerometer may be a triaxial device, although a simpler device may also be used to achieve satisfactory results. As previously stated, in some situations the relevant sensor may be a magnetometer rather than an accelerometer.

[0075] A suitable accelerometer is available from STMicroelectronics under the reference LIS2DTW12, although there are of course many similar alternative devices available from other manufacturers.

[0076] The accelerometer may conveniently be configured to remain in sleep mode when not subjected to a certain level of activity for more than a predetermined duration (in order to reduce power consumption and thus extend battery life in battery-powered devices), in which case the accelerometer may be further configured to wake up quickly when subjected to a certain amount of stimulation - such as an RMS signal value reaching or exceeding a threshold (e.g., along any of the sensing axes), indicating a possible increase in the activity level. The STMicro device mentioned is a "always-on" device, but in sleep mode has a sampling rate of 12.5 Hz, or one sample every 80 ms. Wake-up is fast, and the device can, upon waking, switch to a sampling rate of up to 1600 Hz (although the optional 800 Hz sampling rate is probably plenty fast enough for this application), so one can generally expect that an accelerometer, after waking, will likely continue to detect ongoing stimulation from the wake-up event. This stimulation will result in an analyzable output signal. The STMicro device mentioned (as probably can its counterparts from other manufacturers) can also generate an interrupt signal upon waking, so that the processor can also be notified of the activity.

[0077] The processor 258 may thus be configured to process first sensor data from at least a first sensor, for example from a first sensor such as the accelerometer 275 or a magnetometer, to determine the presence of anomalies represented by the data, and in the event that an anomaly is detected / determined, to generate an output. When a sensing device has a sleep mode, such as the STMicro accelerometer just mentioned, it may be preferable not to consider the fact that the device wakes up as being in itself an abnormal event - although activity subsequent to the device waking up from sleep may very well constitute an abnormal event and should be treated as such by the processor.Thus, an attack event that wakes up a detection device may constitute an abnormal event, even though the mere fact that the detection device wakes up would not in itself be considered an abnormal event.

[0078] The processor 258 may also be arranged to process second sensor data provided by at least one second sensor (e.g., a magnetometer 264 or another accelerometer or magnetometer optionally located elsewhere (i.e., outside the lock and optionally remote from the lock) on the door 102 and / or on the door frame 240) and to consider both the first sensor data and the second sensor data to determine whether the data indicates the existence of an anomaly, and if so, to provide an output. In either case, the output provided by the processor 258 may be transmitted to a remote receiver (such as a user device, e.g., WTRU, a backend system, a remote monitoring station or ARC, a control unit of a security monitoring facility for the premises, or a "smart doorbell" as described elsewhere in this application).Additionally or alternatively, the output provided by the processor 258 may be subjected to a classification process (e.g., using a classifier (or autoencoder) executed on the processor 258 or on a . other processing device in the lock or elsewhere) in order to classify events into particular threat types (e.g. lock punch, lock picking, lock picking, lock impact attack, door impact attack, etc.) and optionally to classify events into particular everyday categories, and optionally again to classify an event as being between a threat and an everyday event.

[0079] Based on such classification, the processor may, for example, be configured to indicate in "tamper detected" alerts and signals the nature of the attack that gave rise to the transmission of the signal. For example, it is desirable to distinguish between signals that result from a mechanical impact from a forced attack on the door or door frame, impacts caused by attempts to pick the lock, and impacts caused by attempts to pry the lock off the door (removal of the lock from its mounting surface). This information may be useful in distinguishing between real attacks (actual intrusions), tampering, and false alarms, i.e., to verify that a security alert can be forwarded to the police for action.

[0080] Optionally, the lock may include a microphone 276, for example arranged to preferentially capture sound from the keyhole and / or from the vicinity of the lock cylinder, and the processor 258 may be arranged to receive signals or data from the microphone as a second sensor or as one of the second sensors. As previously described, the microphone is preferably protected from direct attack through the keyhole, for example by being offset from the keyhole or by being arranged orthogonally to the axis of the keyhole.Optionally, the lock may further or alternatively comprise a pressure sensor 278 and / or an insertion detector 279 (possibly based on the mechanical displacement of a sensing element) for detecting the insertion of an element (such as a key, a strike key or a pick) into an external keyhole of the lock, and the processor 258 (or a remote processor) may be arranged to receive signals or data from any of these sensors as a second sensor or one of the second sensors, or simply as an additional sensor whose signals may contribute to the determination of an abnormal event and possibly its classification. Optionally, the lock may also comprise a microphone for detecting ambient sounds, in particular those coming from the at-risk side of the door.An opening or aperture may be provided in the body of the lock to enable the microphone to detect such ambient sounds, and the opening or aperture may be provided with a grille or similar member to protect the microphone from . attacks. As described above, the microphone is preferably further protected against attack through the opening or hole (e.g., against attack by inserting an object through the opening or hole) by being offset from the opening / hole, or by being mounted orthogonal to an axis of the opening / hole.

[0081] Optionally, the processor is configured to indicate, in the event reporting signal, the identity of the sensor(s) that gave rise to the transmission of the signal, i.e., whether it was the accelerometer 275 or the magnetometer 264 (indicating which magnetometer if the lock contains more than one), the external accelerometer, the external magnetometer, the microphone, the pressure sensor, the displacement detector, or a combination thereof - and, at least in the case where only one magnetometer is provided, the nature of the change in the magnetic environment observed, for example, whether the magnetometer detects a change in the magnetic field from a door magnet 270 or a frame magnet, and whether the change is a reduction in the strength of the magnetic field,an increase in magnetic field strength or a change in polarity – both suggesting a possible attempt to fool the system by adding a malicious magnet to fool the magnetometer (“masking”). This information can also be useful in distinguishing between real intrusion attacks, manipulations and false alarms, for example in distinguishing between vibrations caused by an intrusion and vibrations caused by road or rail traffic, or the impacts of a football.

[0082] Although the foregoing description referred to the context of the electronics of a lock provided in a lock housing together with the main mechanical elements of the lock, the invention also contemplates alternative arrangements in which the electronics of the smart lock are provided in a unit other than the lock housing. For example, with a mortise lock in which the main mechanical parts of the lock are provided within a mortise in a door (preferably contained in a lock housing inside the door), a separate unit may be provided (preferably) on the protected face (e.g., the inner face) of the door to house the electronics of a smart lock, the separate unit being mechanically coupled to the mechanical parts of the lock inside the door.Such an arrangement may be used to enable the functionality of the smart lock to be "upgraded" in a previously installed lock, for example by removing the existing lock cylinder and replacing it with a new cylinder that provides a mechanical connection to the new separate unit by means of which the lock may be remotely locked or unlocked. If the lock assembly provides a latch arrangement, this may also be coupled to the new separate unit to enable a motor or other . actuator (e.g., a solenoid) to actuate both the bolt and the latch of the lock, potentially allowing contactless opening of the door. This approach is perhaps easier to understand with reference to Figure 3.

[0083] [Fig.3A] shows a mortise lock 300 installed in a mortise (cavity) formed within a door 302, as viewed from the protected side of the door, for example its inner side. The lock includes a lock housing 304, shown in phantom, within which are received the bolt 306 and a latch 308. The bolt 306 is shown here retracted, such that only the free end of the bolt is visible in the faceplate 310 of the lock, in a state where the door 302 can be opened from the closed position. The lock 300 is secured within the mortise by screws 312, 312' which pass through the faceplate 310 into the door material. In the illustrated example, the lock is lockable and unlockable by means of the key cylinder 314, shown here outside its bore 315 in the lock towards the internal side (protected side) of the door.The cylinder illustrated here is a Europrofil cylinder in which a key operates a round bolt and tumbler arrangement (other types of cylinders are of course used and the invention is equally applicable to locks using these other types of cylinders, as well as to other locks and lock arrangements such as multi-point and single-point lock arrangements). A keyhole 316 is provided at each end of the conventional cylinder, and by inserting a correctly coded key into a keyhole 316 at either end of the cylinder, it is possible to rotate a rotary cam 318 having an integral tongue 320 which, when installed in the lock, allows the bolt 306 to be moved in and out of the lock case 304, and thus to unlock and lock the door 302 on an associated door frame 240 (not shown).The free end of the bolt enters an opening, or "strike," in the door frame 240 to lock the door in the closed position.

[0084] The cylinder 314 is secured in position in the lock by means of a retaining screw, not shown, which in use passes through an opening 322 in the faceplate of the lock. It should be noted that the lock is shown without the usual cover plate which would normally cover the faceplate 310 of the lock and the screws 322 as well as that of the opening 322. The latch 308 may be opened and closed by a latch drive which engages with a typically square-section opening 324 in the latch (or in a latch drive within the lock). The latch may be coupled to a handle or knob (one on each side of the door) for retracting the latch. Latches are often spring-loaded, so that they return to the indicated position, with the latch tongue popping out of the lock when pressure on the handle or knob is released.The latch is sometimes coupled to be driven by the cylinder, so that additional rotation . of the cylinder in one direction retracts the latch, and rotation in the opposite direction acts conversely - in which case no latch handle may be provided, or a latch handle may be provided only on the protected side of the door.

[0085] In [Fig.3A] two Europrofil cylinders are shown: on the left, a conventional cylinder 314 with a keyhole 316 at each end; on the right, a novel cylinder 314' in which, in place of the keyhole on the protected side of the door, an elongated tongue 326 projects. The tongue 326 will engage with the mechanism of an assembly which includes the mechanism and electronics of an electronic lock. This assembly, which will be described with reference to Figures 3B and 3C, will be mounted in place of the escutcheon 328 which surrounds the bore 315 of the cylinder 314'.

[0086] The modified cylinder 314' is also shown schematically as including a pressure sensor 278 coupled to a bore which communicates with the keyhole on the exposed outer end of the cylinder. The pressure sensor is arranged to detect changes in pressure following insertion of a key (or the like) into the keyhole. Also shown schematically is a displacement sensor 279, which may be provided in addition to or instead of the pressure sensor 278 coupled to a displaceable member which will be moved by insertion of an elongated object (such as a key or a pick) into the keyhole. A microphone 276 may additionally or alternatively be provided at a suitable location on the cylinder 314' to detect sounds of interaction with the lock, e.g., sounds of insertion of a key, a pick, a striking key, etc.

[0087] In [Fig.3B], an assembly comprising the mechanism and electronics of an electronic lock is housed in a knob, or more generally in a handle 330 which is mounted on the door 302 by means of a fixing plate 332 which has itself been mounted on the door in place of the escutcheon 328. The fixing plate 332 is attached to the inner face of the door by means of a pair of fixing screws 334. The fixing plate includes engaging portions, shown here as projections 336, which may be in the form of flanges, which mate and lock into engagement with corresponding formations within the knob 330. The fixing plate 332 and the knob may be held locked together by one or more hidden set screws, each concealed in respective bores 338 through the (here) curved surface of a first part 340 of button 330.In the example shown, the major part of the length of the button forms a body 339 which extends over the first (internal) part 340 and receives it, although other arrangements are of course possible within the scope of the invention.

[0088] A set of one or more visual indicators 341 is also shown in [Fig.3B], for example in the form of RGB LEDs by means of which a state and / or mode or condition of the electronic lock may be indicated. Finally, [Fig. 3B] shows one or more apertures 342 formed in the major surface of the mounting plate 332. Each aperture 342 provides a window through which the magnetic field of a magnet 270 can pass unimpeded. The magnet(s) 270, if used, may be surface mounted on the door within the window provided by an aperture 342, or the mounting plate may be used as a template (or jig) to enable a recess or bore to be formed (e.g., drilled, machined, or cut) in the door structure so that a magnet may be inserted into the door body. If the recess or bore is deep enough, a magnet may be embedded in the door and concealed under a wooden or plastic cap, possibly before mounting the mounting plate to the door.As shown, one or more magnets 270 may be mounted on the door in this manner, in order to cooperate with the magnetometer(s) 264 of the electronic lock as described with reference to [Fig. 2]. It is understood that the position of the opening(s) 342 relative to the engagement portions 336 and the position of the complementary features of the assembly relative to the accelerometer used to detect the magnetic field of the magnet 270 are chosen such that when the assembly is mounted on the door, the magnetometer 264 and the magnet 270 (positioned using the mounting plate used as a template) are aligned for optimum sensitivity.

[0089] [Fig. 3C] schematically shows a longitudinal section through the assembly 330, the mounting plate 332, the lock housing 304, the door 302 and the door frame 240. The assembly includes the knob 330 which is mounted on an inner member 340. The inner member 340 includes the engagement portions 350 which mate and lock into engagement with the engagement portions 336 of the mounting plate 332. As illustrated, approximately half the length of the knob overlaps the inner member 340, the inner member effectively providing an enclosure or housing for the smart lock electronics, although it should be noted that many other configurations are possible within the scope of the invention.The tab 326 of the cylinder 314' extends through an opening in a circuit board 251 (or between a pair of circuit boards) that carries the bulk of the smart lock's electronics, including the processor 258, memory 260, RF transceiver 262, magnetometers 264, and accelerometer 275. As illustrated here, two magnetometers may be used, one for tamper detection using a magnetic field from a door-mounted magnet 270, and the other 264' for detecting the door's open state (and thus potentially intrusion) using a magnetic field from a frame-mounted magnet 266, although the same functionality may be achieved using a single magnetometer. Connected to the processor, but optionally mounted off the printed circuit board 251, are the speaker / buzzer 263 and indicators 341.

[0090] The assembly includes a battery power supply comprising batteries 252, a charging arrangement 254, optionally with a charging socket 256 (preferably on what, in use, will be the underside of the button, so as to be hidden from view under normal circumstances).

[0091] In the example illustrated in [Fig.3C], the locking mechanism comprises an electrically operated actuator 246 (e.g., a motor or solenoid) and optionally an associated mechanism 352, which are configured to together apply torque between the internal part 340 of the knob and the tongue 326, so as to rotate the cylinder 314' in the lock - in order to retract the bolt 244 into the lock housing 304 or to protrude the bolt 244 from the lock housing 304. A force transfer arrangement 354 may be provided between the cam tongue 320 and the bolt. Similarly, the lock may be configured such that rotation (e.g., overrotation) of the cam 318 also actuates the latch 308.

[0092] Although Figure 3 shows suitable arrangements for a conventional door, those skilled in the art will appreciate that the same assembly and modified cylinder could equally well be used in conjunction with a door (or patio door, deck door, etc.) having a multi-point locking system (typically comprising at least 3 locking points which all lock simultaneously when the mechanism is engaged). This system typically operates in conjunction with a cylinder lock (e.g., a Europrofil cylinder lock), although locking typically involves manual operation of a door handle to engage the various locking points with their counterparts before the lock will lock.Although this engagement phase may require the application of too much torque for a sensible application of the described electronic lock addition (if only for the short battery life), the locking and unlocking step separate from the engage / disengage operations could potentially be handled using the system just described.

[0093] In an alternative arrangement, the smart lock is provided with a clutch handle or knob on either side of the door, the smart lock mechanism being configured to control the clutch arrangement so as to allow, in the unlocked condition, retraction of the bolt, and to decouple, in the locked position, the handles / knobs from the bolt such that they are unable to retract the bolt to unlock the door. It is also apparent that rather than providing the additional assembly in the form of a knob, a larger body not in the form of a knob or handle (although it includes potentially a knob or handle to facilitate opening the door) can be used to house the added mechanism - and such an arrangement can be configured to include a motorized latch actuator.

[0094] The smart lock of Figure 3 and its developments as described with reference to Figures 4 to 7 may be designed and configured as an EN certifiable shock sensor for use with EN certified premises security systems (intruder alarm systems) classified as grade 2 or higher (EN50131), including the necessary tamper detection function which will trigger a tamper alert in the event of an attempt to open the device, remove its battery or remove the lock from the door on which it is mounted.

[0095] Figure 4 schematically illustrates methods according to aspects of the invention. In particular, the figure illustrates methods for detecting the presence of anomalies in the signals generated using one or more shock or vibration sensors such as those included in a smart lock as described above. An event is considered abnormal if the corresponding sensor data falls outside the usual range for "normal" use.

[0096] In the method 400, an input is received at 402 from a shock or vibration sensor, which may be a shock or vibration sensor integrated into a smart lock, for example based on an accelerometer or a magnetometer. The received input is then subjected to an anomaly detection process 404. The anomaly detection process 404 may involve both anomaly detection 406 and a classification 410 (optionally according to a two-step process), or may involve anomaly detection 406 but no classification 410 (a one-step process). In the latter case (a one-step process), in the event that an anomaly is detected in the input 402, an output may be provided at 408. If the method includes a classification 410, this may be a further step of a multi-step process (e.g.as a second step after a first anomaly detection step in a two-step process), or it may be implemented in a processing step that involves both anomaly detection and classification. A classification result may be provided as output 412.

[0097] Using a two-step process in which a separate anomaly detection step 406 is implemented before an (optional) classification step, rather than attempting to classify the input 402 directly, provides the advantage that an input 402 that represents an outlier from "usual" behavior can be detected as a potentially suspicious event, even if it is of a type that has not been encountered or modeled before. New types of attacks can thus be detected even if they are not specifically modeled. It may be difficult to classify them subsequently, but at least the system can detect the occurrence of something unusual. This could strengthen the system's sustainability, making it more robust than a single-step classification.

[0098] The anomaly detection process 404 may be based solely on the output provided by a shock or vibration sensing arrangement of the smart lock, such that the output 412 is derived solely from the output provided by the shock or vibration sensing arrangement of the smart lock. Alternatively, the output 412 may be the result of processing the input 402 provided by the shock or vibration sensing arrangement of the smart lock in conjunction with one or more optional additional inputs 414.

[0099] Such optional additional inputs 414 may be provided by one or more sensors optionally provided externally to the smart lock but coupled to the door or door frame, optionally at one or more locations remote from the smart lock. This may be, for example, a sensor such as a magnetic open / close detector, which is adapted to detect the open / close state of the door on which the smart lock is mounted. Similarly, a sensor which is capable of detecting the open / closed state of the door by detecting rotation of the door relative to an axis (which is, for example, capable of encoding the closed, open, and open states, such as in the form -1 / 0 / 1), or which is capable of detecting rotational speed (normalized or not) or angular variation in orientation (e.g., using an accelerometer, magnetometer, or similar sensor).

[0100] An additional input 414 may also be provided by another sensor located in or associated with the smart lock, for example, a microphone located in or coupled to the lock, for example, coupled to a cylinder of the lock, an inductive or capacitive sensor coupled to the lock or cylinder to detect human interaction with the lock (e.g., by contact or proximity), or an insertion detector (such as a microswitch) to detect insertion of a key or pick into the keyhole of the smart lock.

[0101] The anomaly detection 404 or 406 may be implemented using a machine learning algorithm.

[0102] The machine learning algorithm may be trained with training data that includes many examples (such as at least 1,000, at least 5,000, at least 10,000, or more) of sensor output data for “normal” events (such as inserting a key, opening the door, closing the door, unlocking the door, knocking on the door, slamming the door, etc.) and optionally for each class of “abusive” or “attack” behavior (e.g., attempts at lock picking, lock kicking, lock drilling, wrenching, lock / cylinder removal / extraction, or break-in). Training data are labeled according to their “crime” class. Training data for “normal” events may be labeled to indicate “normality,” or may be unlabeled and thus distinguished from data representing “crime” events.

[0103] It is preferable that the training data include data collected in the field rather than simply "lab" data, although it may be necessary to obtain a large amount of "offense" data from attacks conducted on test platforms installed in laboratories or other environments that simulate the real world. The training data need not be a balanced data set, i.e., there need not be the same number of data examples for normal events as for "offense" elements.

[0104] Optionally, for two-stage anomaly detection, the autoencoder may be trained only on normal data or by prioritizing it (omitting and / or penalizing "attack" type data). When the signal is regenerated by the autoencoder, the anomaly detection depends on the autoencoder's ability to associate the original input signal with a "normal" state and to regenerate it. A large error (measured in terms of energy) means that the autoencoder could not correctly associate it with any of the normal states, which implies an anomaly with respect to the "normal" training data. The detection is thus generated. But the "crime" (or "attack") data can be used to train a classifier that is then able to "label" or categorize a crime event - for example, by labeling an event as a lock-picking or lock-punching attack.

[0105] The anomaly detection process can be implemented using unsupervised learning with Gaussian Mixture Models (GMMs) or alternatively using an autoencoder. Autoencoders are neural network-based models that are trained through unsupervised learning to efficiently compress input data to form an encoded representation and then reconstruct the compressed representation to produce an output as close as possible to the original input. An autoencoder can be thought of as a Multi-Layer Perceptron (MLP).

[0106] The classification can be implemented using any suitable classifier, such as a convolutional recurrent neural network (CRNN) architecture which can be based on the combination of an architecture convolutional neural network (CNN) and gated recurrent unit (GRU) architecture.

[0107] [Fig.4B] schematically illustrates the principle of a processor 450 making a determination based on both data received from a sensor 452, for example a shock or vibration sensor such as an accelerometer or a magnetometer, of a smart lock 106, and from an additional sensor 454 which is not part of the smart lock. The processor may also be part of the smart lock.

[0108] The additional sensor 454 may, for example, be a door contact sensor that detects the state (open, closed, opening or closing, degree of rotation, etc.) of the door to which the smart lock 106 is attached, for example, a magnetic sensing-based door contact sensor, such as a magnetometer, a Hall effect sensor, or another magnetically switched device. Alternatively, the additional sensor may be an accelerometer (single-, dual-, or triple-axis) that is not contained within or part of the smart lock, the additional sensor being coupled to the door or door frame in which the door is mounted (e.g., mounted on or in the door or door frame).Alternatively, the additional sensor may be an inductive or capacitive sensor, a pressure sensor, a motion or presence sensor, a microphone, a pressure pad, a radar arrangement, an RF-type presence detection system, a video camera (e.g., the camera of a video doorbell). The processor 450 may implement a classifier and / or an anomaly detection algorithm, optionally based on an artificial intelligence approach. The processor 450 may be configured to process the data from the sensors 452 and 454 to identify the presence of anomalies (e.g., outliers) that may indicate an abnormal event - in response to which the processor may provide an output (e.g., signaling an associated security or surveillance system to flag the existence of an abnormal event).

[0109] A smart lock comprising an accelerometer and / or a magnetometer is thus provided, as well as a unit separate from the lock, mounted on the door or door frame, which itself comprises an additional sensor, such as an accelerometer and / or a magnetometer, and optionally a microphone. Such an arrangement may allow to better distinguish the source / position of a shock or vibration (e.g. on the lock, on the door near or far from the lock, etc.). The provision of differential information could be useful to facilitate the localization of the source of a disturbance, vibration or shock.

[0110] The smart lock (or door sensor) could include a processor to perform the calculations based on the data from both sources, but the data could also be sent to, for example, a control unit of a security monitoring system for the control unit to implement the necessary calculations.

[0111] A shock or vibration sensor may be equipped with a processor configured to process the sensor data to determine the presence of anomalies in the data, and in the event that an anomaly is detected / determined, the processor may be configured to generate an output. In a second optional aspect, the processor may be arranged to analyze the sensor data (or the processed sensor data) to implement a classification for assigning a detected event one of multiple event types, the event types comprising at least one threat class and at least one non-threat class.

[0112] Figure 5A schematically illustrates the elements that constitute an auto-encoder 500.

[0113] Essentially, the auto-encoder comprises an encoder part 502 whose role is to compress the input data to form a 504 encoded representation, and a decoder part whose role is to transform the 504 encoded representation to reproduce as faithfully as possible the original input data. The input data arrives at the input 508 of the encoder and the reconstructed output is provided at the output 510 of the decoder.

[0114] Figure 5B illustrates the layered structure of the autoencoder 500. In this example, the autoencoder comprises 6 fully connected layers. The encoder side applies dimensionality reduction by using fewer nodes for each layer between the input and the encoded portion, for example, going from 32 nodes in layer X0 to 16 nodes in XI and to 8 nodes in X2. The layers of the decoder portion 506 similarly use an increasing number of nodes as one moves away from the encoded representation 504 (or Z), with the number of nodes per layer increasing from 8 to 16 and then to 32. It should be noted that this structure is given only as an example and that embodiments of the invention may use encoders in which the number of layers and the number of nodes in each layer of the encoder and decoder portions differ from those given in this example.

[0115] [Fig.6] schematically illustrates how a trained autoencoder can distinguish between “routine” events – which are labeled here as “in distribution” – and aberrant events (which may represent attacks on the smart lock or the door on which it is mounted), which are labeled here as “out of distribution”. An appropriate threshold t can be selected, for example on the basis of calibration tests, to mark the distinction between “in distribution” and “out of distribution” events. Crossing the threshold can be used to trigger the reporting of an event (more simply the providing an output signal) or to trigger an alarm or other similar response.

[0116] [Fig.7] schematically illustrates how an auto-encoder may be integrated into a smart lock 700 according to one aspect of the invention. The auto-encoder 702, which may correspond to the auto-encoder 500, may be executed on a processor 704, which may correspond to the processor 258. The processor may be a microprocessor but may also be a microcontroller (MCU - Microcontroller Unit) such as a SiliconLabs EFR32xG4 Cortex-M4, which integrates an ARM Cortex-M4 core. A memory 706, which may correspond to the memory 260, stores program instructions for controlling the processor 704, and may in particular store code for implementing the auto-encoder 702 and the classifier 708.The smart lock 700 further includes an RF transceiver 710 that may correspond to the transceiver 262, a power supply 712 corresponding to the power supply 250, and a lock actuator 714 that may correspond to the actuator 246, all of which are operatively connected to the processor 704.

[0117] The first sensor 716 is also coupled to the processor, and may be the accelerometer previously designated by element 275. Optionally, as described above, the smart lock 700 may include at least one second sensor 718, which may take the form of a pressure sensor 278, an insertion detection device 279, or a microphone 276.

[0118] External to the lock 700, one or more second sensors 720, 722 may be provided and configured to communicate (wirelessly or by wired connection) with the processor 704. The second sensor may, as previously described, be a door contact sensor (optionally based on the use of magnetism) that detects the state (open, closed, opening) of the door to which the lock 700 is coupled, or it may be another accelerometer coupled to the door or the door frame in which the door hangs, or another sensor capable of providing data relevant to determining the state of the door and / or the occurrence of an attack or other event, or a combination of these sensors.

[0119] The external sensor(s) 720 / 722 may typically include an RF transceiver for communication with the lock's RF transceiver 710, although the external sensor may be coupled to the processor 704 in another manner (e.g., via a wired connection).

[0120] In one embodiment, the encoder 702 may be executed as a model using, for example, Tensorflow Lite. The models (autoencoder and classifier, if present) may be quantized to int8 from the float32 representation of the weights used during training. Data may be received from each sensor (e.g. 716, 718, 720, 722 or any combination thereof) into packets of 128 samples which are then provided to the input of the encoder program.

[0121] Each packet of 128 samples is then divided into 4 sub-blocks, each consisting of 32 samples, and then the batch of 4 sub-blocks is input into the encoder model. A sample size of 128 was selected in preliminary studies, which revealed that it corresponds to the most appropriate duration for a sampling rate of 200 Hz used by the 714 / 275 accelerometer of the lock. 128 samples give a duration of 640 ms, while 32 samples give a duration of 160 ms. In some embodiments of the invention, these samples and their batches do not overlap and no sliding window is

[0122] applied.

[0123] Tanh activation can be used for the autoencoder neural network, but in other implementations, Relu or leakyRelu can be used instead of Tanh activation.

[0124] In some embodiments, it is possible to create a satisfactory neural network with a total number of weights (sum of the numbers of weights for all layers of the network) of 7624 parameters. This may correspond to a file size, including weight and state, of approximately 17 KB. Such a small model can easily be executed by an MCU, despite the memory constraints typical of MCU implementations. When the sensor is a triaxial accelerometer, the input provided to the autoencoder 702 may be a vector of samples of dimension 3 each, representing the x, y, z coordinates. The format of the sample may be int8, with a length selected from 16, 32, 64 or 128, for example 32.The output of the autoencoder can be a single number that is the detection result, with a length of 1 and an int8 format (having values ​​from 0 to 127) mapped to values ​​from 0 to 100 (i.e., normalized to values ​​within this range). The main mechanism of the model can be to calculate expected values ​​(for normal behavior). Expected inputs should correspond to low values, for example, close to zero with a maximum in practice between 10 and 30 (to allow for some margin of error). “Out-of-distribution” events should give significantly higher values. An alarm threshold can therefore be set between approximately 60 and 90 (in a total range of 0 to 100). The threshold value can be set based on the analysis of the model output values, once the network has been trained on the training data that includes daily and out-of-distribution events.

[0125] In the case where a classifier 708 is used to identify out-of-distribution events, for example to label an abnormal event as a lock picking, a lock punch, etc., a suitable model such as a CRNN architecture model may be chosen to be executed on the processor 704. The classifier may, in the case where the sensor is a triaxial accelerometer, receive an input vector of samples of dimension 3 each, representing x, y, z coordinates. As with the autoencoder, the sample format may be int8, with a length chosen from 16, 32, 64 and 128, for example 32. The output of the classifier may be a vector of the classification result for the labels / classes, optionally with a uint8 format (values ​​0-100) with a length of 12.

[0126] The classification model may be based on supervised learning using deep learning, provided that a well-prepared dataset is available to train the model.With a smaller dataset, it may be useful to use k-fold cross-validation (k=5) to avoid overfitting to specific data and biasing the model towards making bad decisions. Such an approach may still be useful with larger datasets (e.g., 10,000 events or more). It may also be useful to pre-train the classification model on all data for a coupled iteration and later iterate on a properly balanced dataset. Alternatively, one could use a Support Vector Machine (SVM) for some features extracted from the signal. This would be a more classical approach. It has some limitations, including the need to find and select the best features.

[0127] Using unsupervised learning with the autoencoder is, however, likely to yield better results than supervised learning if there is a lack of labeled real-world data - as will typically be the case for at least some attack modes. Unsupervised learning is oriented towards outlier selection, meaning that it should still be able to provide appropriate output when never-before-encountered data is input to the anomaly detection step - such that even new types of attacks can be expected to trigger an alarm (or alert) notification.

[0128] In the case where signals from a second sensor (e.g. an external sensor 720 or 722) are used in addition to those from the first sensor of the lock, the signals from the two (or more) sensors can be combined to provide a long vector (with a wider input). If the second sensor is for example a door status sensor, the lengthened vector could be in the form I door lock accelerometer I status I rotation I.

[0129] Alternatively, the state / rotation data could be represented as a position encoding (the number is sin / cos modulated and added after the first layers), as described in the article available at the following URL:

[0130] https: / / medium.eom / @hunter-j-phillips / positional-encoding-7a93db4109e6

[0131] These signals can then be used as input into the anomaly detection model and / or the classification model, if applicable.

[0132] Alternatively, it is possible to implement in the firmware a decision maker for the detection model, which will only trigger an alarm event when the door on which the smart lock is mounted is closed and there is no rotation of the lock.

[0133] It is possible to generally improve the classification results by using another type of sensor in addition to the data from the smart lock's accelerometer (or magnetometer), because normally the more data points there are, the better the overall classification.

[0134] In practice, there is always a way to normalize the data from different sensors as well as to align them. Adding additional information can generally improve the classification accuracy, for example by 3 to 30% relative. In some particular use cases, however, the classification accuracy can be improved significantly, for example for the detection of a hand on the handle or on the lock (using e.g. capacitive detection), and a key-in-lock detector (using e.g. impedance detection) can help in the classification of particular attack modes such as lock picking, lock picking and lock tapping.

[0135] As previously stated, when using multiple sensors, all sensor data can be combined into a single input, and concatenated to form a vector in the form of a long line. But it is also possible to consider some sensor outputs separately.

[0136] When a microphone is used as a second sensor (either in the lock or as an external sensor), it would make sense to implement either feature extraction, optionally pre-filtering, and possibly a Fast Fourier Transform (FFT), optionally using Mel-Frequency Cepstral Coefficients (MFCC) or Constant Q Cepstral Coefficients (CQCC) to down-rank the data and emphasize only certain frequencies (very low < 50 Hz, probably in the range 2-10 Hz).

[0137] The smart lock 106 may be used as part of a security monitoring system, for example a Grade 2 or higher system according to EN50131, although it may also be used as a standalone component that can communicate with one or more user devices and / or a network-based (e.g., cloud-based) backend - optionally via a local RF connection to a broadband router and from there via a wired and / or wireless connection, or via LPWAN (e.g., via NB-IOT or LTE-M). In one example, such a security monitoring system may be controlled by a controller of a video doorbell such as that identified by the number 100 in [Fig.l], as will now be described.

[0138] [Fig. 8] schematically illustrates the principal constituents of an example of such a video doorbell 100, suitable for use in embodiments of the invention in conjunction with a smart lock as described with reference to Figures 3-7. A video doorbell 100 includes a processor 800, which may be a microcontroller (MCU), with an associated memory 802 that stores program instructions ("software") which, when executed on the processor 400, control the operation of the video doorbell 100. A pair of transceivers 804, 806 may be provided. A first transceiver 804 may support Wi-Fi communication with a Wi-Fi access point (AP), such as the Wi-Fi router 112 of [Fig. 1], for communicating video signals from the video camera 808 and audio signals from the microphone 810.The second transceiver 806 may support a lower bandwidth channel for transmitting control signals to, and event notifications from, alarm devices such as the smart lock 106, external video cameras, door / window contacts (which detect the open or closed status of windows or doors), etc. Typically, the lower bandwidth channel(s) may be provided using one or more appropriately allocated frequencies in the Industrial, Scientific, and Medical (ISM) bands—such as 868 MHz (in Europe). Communications on the lower bandwidth channels are preferably encrypted.The primary reason for using a low bandwidth transmission channel is to enable the use of transceivers (in the alarm devices) having low power consumption (and in particular lower power consumption than the notoriously power-hungry Wi-Fi transceivers) so that the battery-powered alarm devices can achieve the necessary minimum battery life of 3 to 5 years. The transceivers 804 and 806 are coupled to an antenna arrangement 812 that typically includes separate (multiple) antenna elements for both the transceivers 804 and 806.

[0139] One or both of the transceivers 804, 806 may support other transmission bands / protocols, such as BlueTooth (RTM) or Bluetooth Low Energy (BLE) for example. The video doorbell 100 may also support communication via an LPWAN network (e.g., via NB-IOT or LTE-M), optionally including device-to-device (D2D) communication.

[0140] The video doorbell 100 also includes a chime triggering device 814, which may include a mechanical switch or "doorbell push button," and preferably a switch that provides tactile feedback upon activation. As is known, the chime triggering device 814 may be co-located with a lens of the video camera 808, but may also be provided separately such that the lens of the video camera does not have to move when the chime triggering device 814 is activated.

[0141] The video doorbell 100 also preferably includes (or may be connected to) a numeric keypad 816 for entering numbers (such as access codes or PIN codes). The numeric keypad 816 is preferably a mechanical device that provides tactile feedback upon activation.

[0142] The video doorbell 100 also includes a motion detection arrangement 817 that can be used to trigger the video camera for recording the motion-triggered video capture and marking (notifying) the motion - although the video camera may be designed to be always active to provide continuous monitoring. The motion detection arrangement may include one or more passive infrared sensors (e.g., PIR passive infrared sensors or T-MOS thermal sensors). Additionally or alternatively, the motion detection arrangement may use radar and / or ultrasound to determine the distance between an object (e.g., a person) and the video doorbell, and this may be combined with passive infrared presence detection and / or other types of presence detection.The video doorbell may have an associated application or online tool to allow a user to define activity ranges or zones, configure sensitivities, etc., such that notifications regarding motion or presence detection are not, for example, transmitted regarding motion detected in public spaces (e.g., movement of people walking on a sidewalk outside the premises, or of vehicles passing on a road outside the protected premises).

[0143] The video doorbell 100 may also include lighting 818 (in the form of one or more light sources), in particular infrared lighting, but optionally also lighting to provide visible light illumination.

[0144] The video doorbell 100 may also include an optical display 819 for displaying user messages and comments. Although it is possible to include the function of the keypad 816 on the display 819 using a touchscreen, it is preferable to separate these functions and use a separate mechanical keypad 816, both for durability and power consumption reasons.

[0145] The video doorbell 100 may also include a sound output device such as a speaker 819 to reproduce a user's speech on the user device 108, for announcements from the processor 800, and optionally to function as a buzzer or siren during an alarm event - as will be explained later. The presence of at least one sound output device (two or more devices may be provided) and at least one microphone (two or more microphones may be provided) allows for two-way audio communication between a person in front of the video doorbell and another person, for example, a designated user via a user device (e.g., a smartphone with a video doorbell application or a security monitoring system application).The presence of one or more microphones 810 also allows voiceprint identification to be used as a form of access control, such that a premises occupant or trusted visitor (nurse, caregiver, housekeeper, etc.) who has recorded a voiceprint in the system can, for example, enter the home by speaking to the video doorbell, which compares the given voice sample to known recorded voice samples to identify the speaker. If the identified speaker is on a list of approved people, the video doorbell can send an instruction to a smart lock on the relevant access door to allow the identified speaker to open the door to gain access to the protected premises.

[0146] Optionally, the video doorbell 100 also includes a Near Field Communication (NFC) antenna 820 to enable near field communication with an NFC device such as a hardware key (dongle), smart watch, NFC tag, or suitable equipped mobile phone - for example, to authenticate a user and possibly unlock an associated smart lock 106 to gain access to the protected premises.

[0147] The video doorbell 100 is provided with an external housing or enclosure 821 to protect the components of the doorbell. The housing 821 is preferably provided with one or more tamper detection arrangements 813 that generate a tamper alert if the housing is removed from the video doorbell 100. Such a tamper alert is preferably processed by the controller of the surveillance system security (the controller being optionally provided either by the video doorbell itself or by a separate entity), regardless of the armed state of the security surveillance system.

[0148] Users (or owner(s)) of the security monitoring system may be provided with an application by which the security monitoring system controller may be notified of a scheduled event that would otherwise be treated as an alarm event, such that the security monitoring system controller may be configured to effectively ignore the scheduled event when it occurs. For example, it may be necessary to remove the housing 821 from the video doorbell to replace a faulty battery 826. By using the application, a user or service technician may prevent the security monitoring system controller from considering the removal of the housing as a tampering event to be reported as an alarm event. The same principle also applies to the lock 106 and its tamper detection.

[0149] The video doorbell 100 is preferably equipped with a mains power supply 822 coupled to a power supply 824 which includes a rechargeable battery 826 (as an auxiliary power supply for use in the event of a mains power outage) and a charging arrangement 828 for managing the charging of the rechargeable battery 826. As illustrated, the video doorbell 100 preferably also includes an interface 830 (such as a micro USB port, a USB-C port or a “Lightning” jack or the like) for accepting a low voltage power supply (e.g.of 5 V or 20 V, from an external battery or charger) which, in the absence of mains power (or in the event of a mains power failure), can be used to recharge the rechargeable battery 826 on-site, rather than requiring the battery to be removed for recharging, so that the video doorbell can still be powered - which is an important consideration when the video doorbell also functions as a controller of a security monitoring system. Alternatively, the video doorbell can be configured to accept wireless charging (e.g., by induction) from an external power source.

[0150] The function of the video doorbell as a controller of a security surveillance system will now be considered in the context of [Fig.9].

[0151] [Fig.9] schematically illustrates one embodiment of a home security monitoring system (home alarm) 900 designed around a video doorbell 100 that functions as a controller of the system 900. In the illustrated example, the system 900 includes a smart lock 106 as described with reference to FIGS. 2-7, such that the user of the system (e.g., the householder) can remotely unlock the door 102 on which the lock 106 is mounted to admit a visitor after optionally first approving the visitor by means of the video doorbell 100's video camera.

[0152] The system 900 includes alarm peripherals such as a door contact 902, for example on the back door 904 of the house (and optionally also on the front door on which the lock 106 is mounted), which may be a magnet detection switch (such as a reed relay) or a magnetometer or a Hall effect sensor that detects whether the door 904 is open or closed, or an accelerometer-based shock or vibration sensor, and which transmit an alert that is received by the second transceiver 806 (e.g., an 868 MHz transceiver) of the video doorbell 100 and processed by the processor 800 to generate, if the system is armed, an alarm event signal addressed to a remote (e.g., cloud-based) alarm reporting service 110 and transmitted by the first transceiver 804 (e.g., an 868 MHz transceiver) of the video doorbell 100.Wi-Fi); the video doorbell speaker can also be immediately activated as a siren or to provide another audible deterrent. The transmitted alarm event signal may be received by the premises Wi-Fi access point (AP) 112 and thereby transmitted via a broadband connection over the Internet 115 to the cloud alarm reporting service 110 (or may be transmitted via LTE-M, NBIoT, or otherwise). If a door sensor 902 is provided for the door on which the lock 106 is mounted, the sensor 902, which may be accelerometer-based or magnetic field-sensitive, may function as a second or one of the second sensors 454 and / or be the source of the optional additional input 414. An alarm event signal may be transmitted in the event that the smart lock 106 determines the existence of an abnormal event (as described with reference to any of FIGS. 4-7).The smart lock may optionally determine the existence of an abnormal event, even in the case where the door sensor 902 does not itself detect a stimulus corresponding to its own alarm threshold: the alarm event may be reported by the smart lock to the video doorbell for its subsequent transmission (although a transmission from the smart lock 106 directly to the alarm reporting service 110 rather than via the video doorbell is also envisaged).

[0153] In the cloud alarm reporting service 110, the identity of the security monitoring system (which may be that of the video doorbell 100 or the smart lock 106) that is included in the alarm event signal may be checked against a database of registered system identities to retrieve an address (e.g., a Session Initiation Protocol (SIP) address, a mobile phone number, an email address) corresponding to the registered alarm system identity. The alarm reporting service 110 may then send a notification to the registered address such that a user receives, for example, a notification (e.g., a native notification) on their smartphone 108 or other personal electronic device, and / or an automated phone call providing details about the alarm event - e.g., the identity / type / location of the sensor that was triggered (e.g., the back door contact sensor or the smart lock 106). A user receiving such a notification may be able to request video streaming or image transmission from cameras 906, 906' installed in or on the premises (e.g.,a home) of the security monitoring system installation: for example, a push notification may include one or more “action buttons” that, upon activation by the user, cause the user’s device 108 to transmit an instruction to the alarm reporting service 110 so that the alarm reporting service 110 sends an instruction to the security monitoring system controller 900 (i.e., the video doorbell processor) to transmit activation signals intended to wake up the cameras 906, 906”, causing them to capture and send images and / or video.The security monitoring system controller 900 (i.e., the video doorbell processor) may receive these video transmissions using the first (Wi-Fi) transceiver 804, and the same transceiver may be used to then transmit the images / videos to the alarm reporting service 110 using the access point 112 and the Internet 115. The alarm reporting service 110 may in turn transmit the images / videos to the user's device 108. The alarm reporting service 110 may be configured to store (if the user has paid the necessary subscription fees) images / videos captured in this manner.

[0154] Each of the video cameras 906, 906' may in fact be an "off-the-shelf" video camera (in order to reduce costs) that includes only one transceiver, and this transceiver may support Wi-Fi but not a secondary low-bandwidth channel. In such cases, the video doorbell as the system controller of the security monitoring station must communicate with and receive communications from the cameras only by means of Wi-Fi, i.e., using the first transceiver 804. It is preferable, however, that the video cameras 906, 906' support both Wi-Fi and a suitable second communication channel with the second transceiver 806 of the video doorbell - and this functionality should preferably be provided by the inclusion of a second transceiver in each of the cameras 906, 906'.Optionally, the video cameras each support an LPWAN technology such as LTE-M and / or NB-IoT.

[0155] The user's device may also be programmed with a companion application that allows the user to send instructions to the video doorbell (which, in this embodiment, functions as the controller of the system security monitoring) to control the activation of selected video cameras so that the user can "keep an eye" on the house when he is away, and also to arm / disarm the alarm system. Images and videos from the cameras 906, 906' may be transmitted via Wi-Fi between the camera 906, 906' and the video doorbell 100 - and these images and videos will therefore be received by the first transceiver 804, which is also used for onward transmission to the access point 112. In some embodiments however, the system may optionally also be configured to send lower resolution images and / or lower frame rate videos using the low bandwidth channel (e.g. 868 MHz) received by the second transceiver, to serve as a backup solution in case of disruption of the high bandwidth channel (e.g.Wi-Fi), the video doorbell processor being configured to transmit the first image / video that arrives, and then to transmit the corresponding image / video of higher resolution / frame rate if and when it arrives. Similarly, the cameras may be arranged to use an LPWAN such as LTE-M and / or NB-IoT for the transmission of video images and optionally duplicated images of different resolutions and / or frame rates as described above.

[0156] Similarly, notifications are also preferably sent to the alarm reporting service 110 (and subsequently to the user device 108) regarding alerts received from the smart lock 106 (e.g., upon determining the existence of an abnormal event) or other alarm devices such as motion-detecting cameras 906, 906' and a window contact sensor (which may also or alternatively function as a window shock or vibration sensor) 908 that is associated with a window 810.

[0157] Similarly, an anomaly report or tamper alert received from the lock 106, for example, following an attempt to strike the lock, to pry the lock 106 from the door 102, or following a detection of the door being opened by a magnetometer of the lock while the lock has not been unlocked, or following the detection of an abnormal event (such as an attempt to pick the lock or another attack - such as the detection of a drilling attack - on the lock), may be detected and reported by the processor of the lock based on the signals from the accelerometer 275 of the smart lock (e.g., as described with reference to any of FIGS. 4 to 7), and may be processed by the processor 800 of the video doorbell (in this embodiment) and thus be notified to the user via the alarm reporting service 110.Event notifications from alarm devices can generally be provided over a low bandwidth channel for reception by the second transceiver 806 of the video doorbell, since most simpler devices (e.g., door contacts and . motion, shock or vibration sensors) do not require a high bandwidth channel (they do not require Wi-Fi for example) but must be inexpensive and have a very long battery life - such that they will typically each contain only a single transceiver, configured to communicate with the second 806 transceiver (both LTEM and NB-IoT also allow low power consumption, especially for low data rates, and support long-distance communication, which also makes them attractive for this application).As already mentioned, alarm devices in the form of video cameras can benefit from a high-bandwidth channel, such as Wi-Fi, because the fast transmission of good quality video images (in color, with high resolution and a sufficiently high frame rate) usually requires quite high bandwidth, and therefore video camera devices almost always include a Wi-Fi transceiver and are therefore almost always powered primarily by the mains (preferably with a backup battery).While the Wi-Fi transceiver could be used for the transmission of event notification signals - to be received by the video doorbell using the first transceiver 804, it is preferable (budget permitting) to provide video cameras that can report events using a low bandwidth channel (for reception by the second transceiver 806) and preferably include an additional transceiver to enable this (although of course only one multi-function transceiver can be used). By providing the video camera peripherals with both narrowband and wideband communication capability, we make it easier and more convenient to upgrade a system such as that of [Fig. 9] to the enhanced system of Fig. 10, as described below.

[0158] The ability of the video doorbell to detect presence (e.g., using one or more PIR sensors, a TMOS sensor, Wi-Fi detection, radar, or the camera) allows the video doorbell to function itself as an alarm device for the system. In this way, when the system is armed, the video doorbell can communicate an alert just as it does when another alarm device is triggered. While this may only be appropriate in certain circumstances (such as those dictated by the likelihood of false alarms caused by "safe" actors), it is potentially an attractive option.

[0159] The processor 800 of the video doorbell 100 in its form as a controller of the security monitoring system may also be configured to check the status of the peripherals, including the quality of the communication channel, the status of the battery of the peripheral, etc.

[0160] It is contemplated that a video doorbell having the functionality of the security monitoring system controller may not only be installed as a retrofit in existing homes, but also in new homes, optionally with the smart lock 106 as described with reference to any one of Figures 2 to 7, together with at least one alarm device, optionally in the form of a motion detection video camera, 906, 906', and preferably a further alarm device in the form of a door contact for detecting the state (open or closed) of the front door (provided with the smart lock 106) or the back door (or other secondary access door to the premises).Such a system could be installed at relatively low cost, and the smart lock 106 and its door magnet(s) 270 (and optionally its frame magnet(s) 266) could, for example, be factory-installed - or at least the door could be factory-prepared for on-site installation of the smart lock 106 and door (and frame) magnet(s) 266 and 270. The first occupant of the house would then benefit from the enhanced security provided by a professionally installed alarm, although it would only have a subset of all the possible features of such a system - and in particular, it would not have a professional monitoring service.

[0161] The design and configuration of the video doorbell and alarm peripherals - including the smart lock 106 - are such that once a security monitoring system based on their use, such as that schematically illustrated in [Fig. 9], has been installed, upgrading the system to include professional monitoring by means of a remote alarm receiving center (ARC) can be achieved relatively easily. Such an upgrade may of course also be of interest to an occupant who has already installed a system such as that illustrated in [Fig. 9]. This upgrade will now be described with reference to Figure 10.

[0162] Figure 10 corresponds generally to [Fig. 9], as it is based on the assumption that a system such as that illustrated in [Fig. 9] has been upgraded to include the benefits of professional monitoring by means of a remote alarm receiving center 1002 - although it is also envisaged that security monitoring systems such as that illustrated in Figure 10 may be installed from the outset. The security monitoring system 1000 differs from the system 900 of [Fig. 9] in that it includes a standalone alarm control unit 1004 which is configured to control the security monitoring system 1000, rather than depending on the controller functionality provided by the video doorbell. The video doorbell 1006 of the system 1000 may be, and preferably is, the video doorbell 100 which controlled the system 900, but with its processor reconfigured (e.g.through proper reprogramming or system updates) to function (at least in part). as a peripheral reporting to the alarm control unit 1004 rather than continuing to act as a controller of the system 900.

[0163] The present invention thus relates, as an alternative, to an initial installation which comprises a video doorbell as described above, functioning as a controller of a security monitoring system which preferably includes a lock, such as the lock described above with reference to any one of Figures 2 to 7, and at least one further installed alarm device comprising either a door / window contact for detecting the state of a door or window, or a door and / or window shock or vibration sensor for detecting physical attacks against the door or window concerned, or a separate motion-detecting video camera, the alarm device comprising a transceiver for communicating with the first transceiver of the video doorbell.The present invention further relates to a method comprising upgrading such an initial installation by introducing a standalone alarm control unit 1004 which is configured to control the security monitoring system 1000, and reconfiguring the video doorbell such that its processor is reconfigured to function (at least in part) as a peripheral reporting to the alarm control unit 1004 rather than continuing to act as the system controller 900. Once the system is upgraded, the video doorbell may serve as the system control interface, such that a user (e.g., an occupant) may send arm / disarm and lock / unlock commands therefrom which are received and processed by the standalone alarm control unit 1004 which serves as the system controller.

[0164] The alarm control unit 1004 includes, like the video doorbell described with reference to [Fig. 8], a processor 1010, which may be a microcontroller having an associated memory 1012 that stores program instructions ("software") that, when executed on the processor 1010, control the operation of the alarm control unit 1004 and the security monitoring system 1000. At least one pair of transceivers 1014, 1016 may be provided. A first of the transceivers 1014 may support Wi-Fi communication, wherein the alarm control unit is configured to operate as an access point of a Wi-Fi network (having a first SSID that is dedicated to the security monitoring system) that may operate in parallel with another Wi-Fi network (having a second SSID) that may support home Wi-Fi needs such as supporting video streaming services (e.g.Netflix, Amazon Prime, RTM), working from home, etc. The alarm control unit 1004 is configured to receive video signals from video cameras 906, 906' using the first of the transceivers 1014. The second transceiver 916 can . support a lower bandwidth channel for transmitting control signals to, and event notifications from, alarm devices such as the 1006 Video Doorbell, 106 Smart Lock, 906, 906' Video Cameras (if they have the appropriate functionality, otherwise event notifications, etc. are received via Wi-Fi), Door / Window Contacts such as 902 and 908, etc.

[0165] The low bandwidth channel(s) may typically be provided by using an appropriately allocated frequency in the Industrial, Scientific and Medical (ISM) bands - such as 868 MHz (in Europe).

[0166] Communications over low bandwidth channels are preferably encrypted. The primary reason for using a low bandwidth transmission channel is to enable the use of transceivers (in alarm devices) with low power consumption (and in particular lower power consumption than notoriously power-hungry Wi-Fi transceivers) so that battery-powered alarm devices can achieve the necessary minimum battery life of 3 to 5 years. The transceivers 1014 and 1016 are coupled to a suitable antenna arrangement (not shown) which typically includes separate (multiple) antenna elements for the two transceivers 1014 and 1016. One or both of the transceivers 1014, 1016 may support other transmission bands / protocols, such as BlueTooth (RTM) or BLE, etc.At least one transceiver is provided, which is preferably configured or configurable to communicate using 4G or 5G (or using any other suitable PLMN data protocol) to communicate with PARC 1002, for example in the event of a loss of power at the premises or loss of wired connection to the Internet 115—for example, as a result of malicious action by a wrongdoer.

[0167] The alarm control unit 1004 is preferably configured to operate at least in a "remote armed" mode, in which the perimeter security of the premises is monitored by means of door and window sensors such as the door sensor 902, the smart lock 106 and window sensors 908, and in which at least one presence sensor monitors the interior of the premises. The internal presence sensor may for example be a passive infrared sensor (or T-MOS sensor) which may or may not be associated with a video camera. Additionally or alternatively, an internal presence may be detected using so-called Wi-Fi detection, where the alarm control unit executes a Wi-Fi detection algorithm and the video doorbell and / or another static Wi-Fi node (or access point) acts as an illuminator for the Wi-Fi detection.The alarm control unit 1004 is preferably also configured to operate in a disarmed mode, in which an internal presence is not monitored and . wherein alert signals received from the door and window sensors are largely ignored by the alarm control unit 604. But the alarm control unit 1004 is preferably configured to respond to tampering signals from, for example, the lock 106 (and also to reports of determination of an abnormal event), or other alarm peripherals, even in the disarmed mode - because thieves and other bad actors may take advantage of the disarmed state to attempt to sabotage or inactivate peripherals and other components of the system 1000.The alarm control unit 1004 is preferably also configured to operate in at least one "armed at home" mode, in which internal presence is generally not monitored (or at least alerts received regarding a detected presence are ignored by the alarm control unit 1004), but in which the security of the perimeter or part of the perimeter is monitored, such that event signals relating to detected door and window openings are alerted and possibly reported as alarm events (optionally after allowing time for an on-site user to mark the detected event as a false alarm, e.g., by entering a security code on a user interface, or by presenting a dongle, token or mobile phone, e.g., on an NFC interface).

[0168] The alarm control unit 1004 is operatively coupled to a remote alarm receiving center (ARC) 1002, preferably by both a wired (e.g., broadband) connection and one or more wireless connections (e.g., SigFox (RTM), PLMN using, for example, a GPRS, LTE, 4G, or 5G protocol) - for example, using either NB-IOT (narrowband IoT supported by 3GPP) and LTE-M (LTE machine type communication) protocols. Notifications received by the alarm control unit 1004 that match the profile of potential "alarm events" (based on one or more of the identity of the sensor providing the notification, the nature of the notification, the arming status of the alarm system 1000, the time of day, the day of the week, etc.) may be reported to PARC 1002 along with the identity of the alarm installation 1000.Such a notification may be generated or caused by the smart lock 106 detecting or determining that an abnormal event has occurred. If the notification is associated with images or video, for example, because the triggered sensor was a camera or was associated with a camera, then the images / videos may also be sent to PARC. PARC's human operators typically review the notified event and attempt to verify that it is indeed an alarm activation event, based on, for example, the content of the images / videos, sounds received from microphones on the premises, interactions with a user on the premises (e.g., via a user interface in . the monitored premises). If a notified event is verified as an activation event (e.g., a break-in, burglary, fire, etc., or a medical emergency), an alert is sent to local law enforcement and / or other emergency response personnel / services (e.g., fire department, EMS, etc.). The ARC may further contact a registered user or other human contact (parent, support person, caregiver, etc.) by sending a push (or other) notification to a user device 108, for example, using a technique as described previously.

[0169] In a system as illustrated in Figure 10, the video doorbell 1006 may communicate events to and receive control signals from the alarm control unit 1004, preferably using a low bandwidth channel via the second transceiver 1016. The wider bandwidth of a Wi-Fi channel, provided by the first transceiver, is preferably used for transmission of images / videos from the video doorbell to the control unit 1004, the control unit then transmitting the images / videos to PARC 1002 and optionally to a user device 108 via a backend service 110 that sends push notifications to a personal communication device 108 of the user when the doorbell is operated (or optionally when someone approaches the doorbell 100).To support such communications, the control unit 1004 may use its broadband connection 114 to the Internet 115, with the backend service 110 sending push notifications to a public land mobile network (PLMN) 116 by which the notifications are delivered to the user's personal communication device 108. The same communication path may be used to stream live video (and audio) from the video doorbell 100 to the user. A reverse communication path may also be supported to allow the user to speak to a person at the doorbell 100, and also to allow the user to activate (i.e., unlock) the lock 106 to allow the person at the door to open the door 102. The lock activation (unlock) signals are transmitted to the lock 106 via the control unit 1004 which communicates with the lock 106.

[0170] The video camera, processor, and video doorbell software are preferably configured together to capture and stream (on demand or as needed) high definition (e.g., 1080p) video.

[0171] The system controller, whether provided by the video doorbell 100 or the alarm control unit 1004, is preferably configured to respond to entry of a "duress code" on the keypad of the video doorbell by alerting either PARC 1002, in the case where the controller is the control unit 904, or by alerting designated persons with calls (e.g., SIP calls) or notifications, preferably in each case providing captured images or video to the time of duress code entry (and perhaps 30 seconds or more before the duress code is entered). A duress code is a code (e.g., a PIN) that can be used to disarm the security monitoring system, or to gain access to the premises, or both, but is in addition to a "normal" access code.

[0172] The idea is that under normal circumstances, a normal code will be used, and the use of such a code does not give rise to any special notification. But in the event that a user is threatened (e.g., by an abusive partner or ex-partner, a thief, or any other type of wrongdoer) and forced to enter the protected premises, the user enters the duress code rather than the normal code, thereby signaling the existence of a duress condition. Upon receipt of a duress notification, PARC officers may contact the police or other security personnel to enable intervention. Similarly, if the duress notification is received by a friend, neighbor, or relative (e.g., in the case where the video doorbell acts as a controller for the security surveillance system), that person may intervene or involve the police to intervene.

[0173] The system controller, whether provided by the video doorbell 100 or by the alarm control unit 604, may be configured to allow access to the protected premises (e.g., by unlocking the lock 106 and either disarming the system or at least temporarily ignoring a door open signal relating to the relevant access door) based on a facial identifier - i.e., by recognizing the face of a visitor as matching the face of a stored identity (the controller storing, or having access to a memory where is stored, facial recognition data for one or more known identities to which access is to be provided).

[0174] The system controller, whether provided by the video doorbell 100 or by the alarm control unit 1004, may be configured to allow access to the protected premises (e.g., by unlocking the lock 106 and either disarming the system or at least temporarily ignoring a door open signal for the relevant access door) based on recognition of a QR code, e.g., a dynamic QR code, that is presented to the video doorbell camera—e.g., by a visitor (or occupant) displaying a QR code on the screen of the phone or other wearable device (such as a smart watch). Any QR code generated by, or in association with, the system (e.g.,by a suitable backend system or by the user's video doorbell or system application) preferably has a limited lifetime of 24 hours or less (shorter lifetimes, e.g., 12 hours, 6 hours, or 4 hours, can also be used advantageously) in order to enhance system security.

[0175] It should be noted that the preceding description simply illustrates examples of the invention without, however, limiting its scope.

Claims

Claims

1. A vibration sensor for a door or window, configured to detect an intrusion, the vibration sensor comprising a processor configured to apply a machine learning model to the processing of sensor data to determine the presence of anomalies in the data and, in the event that an anomaly is determined, to generate an output.

2. A vibration sensor according to claim 1, wherein the sensor comprises an accelerometer or a magnetometer.

3. A vibration sensor according to claim 1 or 2, wherein the processor is programmed to operate as an autoencoder to process the sensor data.

4. The vibration sensor of claim 3, wherein the autoencoder is configured to adjust and / or classify a received signal based on pre-trained non-anomalous signal data and to re-estimate the resulting signal, and the processor is configured to detect an anomaly based on a difference between the received signal and the re-estimated signal.

5. A vibration sensor according to claim 4, wherein the processor is configured to detect an anomaly based on an energy difference between the received signal and the re-estimated signal.

6. A vibration sensor according to any preceding claim, further comprising a microphone coupled to the processor.

7. A vibration sensor according to any preceding claim, wherein, in the event that the processor determines that an anomaly exists, the processor is programmed to analyze the sensor data, or processed sensor data, to associate a detected event with one of multiple event types, the event types comprising at least one threat class and at least one non-threat class.

8. The vibration sensor of claim 7, wherein the threat class comprises at least one of lock picking, lock tapping, and lock picking.

9. An electronically controlled lock comprising a vibration sensor according to one of the preceding claims, and optionally wherein the lock comprises a driver intelligent electronic lock functionally coupled to a mechanical lock.

10. An electronically controlled lock according to claim 9, wherein the lock comprises a keyhole and the machine learning model is trained to recognize as abnormal events at least one of lock picking, lock tapping and lock wrenching.

11. An electronically controlled lock according to claim 10, wherein the lock comprises a microphone coupled to the keyhole, and the processor is arranged to process signal data from the microphone to facilitate recognition of abnormal events.

12. An electronically controlled lock comprising a vibration sensor according to claim 2, wherein the processor is configured to implement an autoencoder, the autoencoder is configured to classify a signal received from the accelerometer based on pre-trained non-anomalous signal data and to re-estimate the signal accordingly, and the processor is further configured to detect an anomaly based on a difference between the received signal and the re-estimated signal.

13. An electronically controlled lock according to claim 12, wherein the processor is configured to detect an anomaly based on an energy difference between the received signal and the re-estimated signal.

14. An electronically controlled lock comprising a vibration sensor according to claim 2, wherein the processor is configured to execute a first classifier to detect anomalies in the signals received from the accelerometer or magnetometer relating to an event and, upon detection of an anomaly, to execute a second classifier to attempt to classify the event into one of multiple event classes, at least one of the classes corresponding to an alarm event, and at least one of the classes corresponding to a non-alarm event, the processor being configured to signal an alarm event in the event that the classifier determines that an event is an alarm event, and optionally also in the event that an event cannot be classified as a non-alarm event.

15. A premises security monitoring installation comprising at least one vibration sensor according to any one of claims 1 to 8, and / or an electronically controlled lock according to any one of claims 9 to 14.

16. A premises security monitoring installation, optionally according to claim 15, comprising a control unit for controlling the operation of the security monitoring installation, the installation comprising: a window or door vibration sensor having an accelerometer or magnetometer incorporated therein; an electronically controlled lock, optionally according to any one of claims 9 to 14, having an accelerometer or magnetometer incorporated therein; the window or door vibration sensor and the smart lock being configured to transmit sensor signal data from the at least one accelerometer / magnetometer to the control unit;and the control unit being configured / programmed to generate a score based on the signal data received from the one or more accelerometers / magnetometers, optionally wherein the control unit comprises a processor that is configured to apply a machine learning model, such as an autoencoder, to process the received signal data so as to determine the presence of anomalies in the received signal data.;

17. A premises security monitoring installation, optionally according to claim 15 or 16, comprising a control unit for controlling operation of the security monitoring installation, the installation comprising: an electronically controlled lock, optionally according to any one of claims 9 to 14, coupled to a door of the premises, an accelerometer or magnetometer being incorporated in the electronically controlled lock; a door vibration sensor having an accelerometer or magnetometer incorporated therein, coupled to the same door at a position remote from the lock; the door vibration sensor and the smart lock being configured to transmit sensor signal data from the at least one accelerometer / magnetometer to the control unit; and the control unit being configured / programmed to generate a score based on the signal data received from the one or more accelerometers / magnetometers, optionally wherein the control unit comprises a processor configured to apply a machine learning model, such as an autoencoder, to process the received signal data so as to determine the presence of anomalies in the received signal data.

Images