Cloud-based, on-demand object recognition system to detect threats and issue real-time alerts when active

Abstract

A system includes normally 'dormant' artificially intelligent object recognition algorithms trained to detect threats in streams, images or frames from onsite network-connected security cameras, the normally 'dormant' artificially intelligent object recognition algorithms running, for example, on an offsite network-connected graphics processing unit or units. The effectiveness of surveillance, security, or law enforcement response to threats is improved by enabling the activation, via network-connected device, of the normally 'dormant' artificially intelligent object recognition algorithms, only when a threat is suspected or detected or an incident is already underway. The system is able to monitor in real-time where weapons, persons or vehicles of interest are across all or a subset of the onsite network-connected security cameras.

Description

CLOUD-BASED, ON-DEMAND OBJECT RECOGNITION SYSTEMTO DETECT THREATS AND ISSUE REAL-TIME ALERTS WHEN ACTIVECROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of priority to US provisional application serial no. 63 / 618,131 filed on January 5, 2024, and to US provisional application serial no. 63 / 645,736 filed on May 10, 2024. The priority applications are incorporated herein by reference for all and any purposes.FIELD

[0002] This disclosure relates generally to processes, and systems for improving the effectiveness of surveillance, security, or law enforcement response to threats by enabling the activation, for example via mobile device, of normally ‘dormant’ artificially intelligent (Al) object recognition algorithms trained to detect threats in streams, images or frames typically from onsite network- connected security cameras, the algorithms typically running on an offsite network-connected graphics processing unit or units. These processes and systems may also be capable of issuing alerts to other network-connected devices. The threats the algorithms are able to detect may include, but are not limited to weapons, faces and characteristics of individuals, and features and license plates of vehicles.BACKGROUND

[0003] According to the Center for Homeland Defense there were, from 1999-2021, roughly 292,000 students at 310 schools that experienced gun violence, with 157 fatalities and 365 injured. Additionally, based on statistics from the FBI, the number of total active shooter incidents has more than doubled in recent years, up to 61 in 2021 from just 30 in 2019. These occurrences are increasing, while solutions seemingly lag behind.

[0004] While it’s not a secret that crime beyond mass attacks is also an issue, there is clear data to support this public sentiment. According to the FBI, in 2019 in the US alone, there were an estimated 21 1 ,1 13 robberies, 612,187 thefts of motor vehicles, 1,525,817 thefts from motorvehicles, 1,309,719 instances of shoplifting and theft from a building, and 714,621 instances of aggravated assault.

[0005] The most common, currently available solutions to prevent and respond to mass attacks and other serious crimes are standard CCTV cameras, which may or may not have ‘always active’ threat detection algorithms running, and onsite or offsite law enforcement or security officers.

[0006] The issue with standard CCTV cameras that don’t have ‘always active’ threat detection algorithms running, is that they are generally only helpful for forensic purposes after a crime has been committed. On large and sprawling campuses and in large buildings or complexes, it is cost- prohibitive to have individuals watching all cameras, at all times, for all threats, and therefore law enforcement and security personnel generally only becomes aware of an incident after an individual has seen a crime in progress or after someone has already been harmed.

[0007] Meanwhile CCTV cameras that have ‘always active’ threat detection algorithms are extremely useful, as active shooters often reveal their weapon in plain view of a camera minutes before they start shooting, enabling law enforcement to respond meaningfully faster, and to save lives. Additionally, faces and vehicles of threats are often known beforehand, and these algorithms can enable staff, security and law enforcement to know immediately when a person or vehicle that is a known or suspected threat enters their domain, enabling the prevention of a crime whereas otherwise, the crime would essentially inevitably occur.

[0008] The main issue with currently available options for threat detection algorithms, is that based on currently available commercial options there is no mechanism or means for individuals or users to be able to quickly and easily scale up and scale down threat detection algorithm functionality from a network-connected device, there only currently exist highly manual and time-intensive options to activate or deactivate threat detection algorithms on cameras. The graphics processing units or cloud computing services required to run these algorithms are expensive. While is it worth the investment to have these algorithms running on cameras that are highly likely to spot a threat first, like entrance points, high-traffic areas or perimeter cameras, it is cost-prohibitive for most customers to have these algorithms running on all cameras at all times, even if having these algorithms running on interior cameras would be extremely helpful if a threat has been madeagainst the facility, a threat has been detected by one of the perimeter cameras, or a serious incident has started in the area.

[0009] In summary, CCTV cameras without threat detection algorithms are an affordable solution, but are largely ineffective in stopping threats quickly or before they begin. And while running threat detection algorithms at all times on most cameras is an effective solution to stop threats quickly or before they begin, it is expensive and cost-prohibitive for most customers. Thus, there is a continuing need in the art for processes, methods and apparatus for improving the effectiveness of surveillance, security, and law enforcement response to threats by enabling the activation, via network-connected device, of normally ‘dormant’ artificially intelligent object recognition algorithms trained to detect threats in streams, images or frames from onsite network-connected security cameras, the normally ‘dormant’ artificially intelligent object recognition algorithms running on an offsite network-connected graphics processing unit or units.SUMMARY

[0010] The processes and systems disclosed herein involve a network-connected mobile device or a network-connected camera and processing unit, which may be onsite or offsite, that can activate one or more normally ‘dormant’ artificially intelligent object recognition algorithms (also referred to herein as the threat recognition algorithms) running on an offsite network-connected graphics processing unit or units. The algorithms are trained to detect threats in streams, images or frames typically from onsite network-connected security cameras. These processes and systems may also be capable of issuing alerts, audible or otherwise, to other network-connected devices, mobile or otherwise, for example. The threats the algorithms are able to detect may include, but are not limited to weapons, faces and characteristics of individuals, and features and license plates of vehicles.

[0011] In some embodiments, the network-connected security cameras that are capable of feeding the threat recognition algorithms their streams, images or frames are Internet Protocol (“IP”) onsite cameras. These IP onsite cameras may have the ability to provide their streams or data through Real-Time Streaming Protocol (“RTSP”) or Open Network Video Interface Forum (“ONVIF”) streaming links. In some embodiments, the network-connected security cameras are connected tothe network communication infrastructure wirelessly via Wi-Fi, cellular and / or satellite connection and / or may be powered wirelessly by a battery pack that may or may not be accompanied by a solar panel or other off-grid power source. The camera the network-connected security may also be powered by standard wired power sources and / or connected to the network by standard wired connection sources.

[0012] In some embodiments, the network-connected security cameras use laser and / or nearinfrared (NIR) optics for enhanced threat recognition, such as detection of persons, vehicles and weapons at long distances (over 100m).

[0013] Preferably, when users or other authorized persons identify a threat, they are able to activate the previously ‘dormant’ threat recognition algorithm through an application, text message, phone call or other means on their network-connected device, which may be, but is not limited to, an iOS or Android mobile device. In order to activate the previously ‘dormant’ algorithm, the application on the mobile device may use a single button or buttons within the application window or on the device screen, or receive a simple command (e g., a vocal command) that activates the algorithm. For example, the previously ‘dormant’ algorithm may be activated with a text message or a phone call. In other embodiments, the previously ‘dormant’ algorithm may be alternatively or additionally activated after a threat is detected by an already active (e.g., an always active) threat recognition algorithm running on an onsite or offsite network-connected graphics processing unit or units. Preferably, the previously ‘dormant’ algorithm is able to be very efficiently activated in 60 seconds or less by most adults from a network-connected device. Thus, the activation of the previously ‘dormant’ algorithm for threat detection can occur manually, at the command of the user or other authorized person with the network-connected device, mobile or otherwise, or, automatically, at the command of a threat recognition algorithm running on a processing unit actively detecting threats from data of a network-connected camera. In some embodiments, all cameras in the system only record and analyze images or streams when motion is detected. Therefore the ‘active’ Al object recognition algorithms will generally process the images or streams any time motion is detected by the associated onsite camera(s), and the ‘dormant’ Al object recognition algorithms will process the images or streams any time motion is detected by the associated cluster or set of cameras when the ‘dormant’ Al object recognition algorithms have been activated.

[0014] In some embodiments, within the application running on and / or integrated with the network-connected device, there is a single button or simple way to start tracking a person’s face or other characteristics or a vehicle associated with a threat, after a weapon or other threatening object is identified by activating previously ‘dormant’ threat recognition algorithm.

[0015] In some embodiments, the activation of the previously ‘dormant’ threat recognition algorithm uses a network-connected microphone or audio sensor and a processing unit that has an algorithm trained to identify, for example, a gunshot. In particular, the network-connected microphone or audio sensor, as well as cameras, may be integrated in a set of onsite network- connected 'dormant' devices that were previously not actively detecting. The set of network- connected 'dormant' devices may be previously determined by a user or other authorized person, or developer.

[0016] After the previously ‘dormant’ Al object recognition algorithm is activated by a network- connected device or an already active Al object recognition algorithm, the streams, images or frames from onsite network-connected cameras are sent through the network communication infrastructure, typically to an offsite network-connected graphics processing unit or unit, which may be, but is not limited to, a private server, a shared server, or processing services or graphics processing units provided by cloud computing services like Amazon Web Services, Microsoft Azure, Google Cloud, or other services of the like.

[0017] In some embodiments, the alert that is issued by the system, once the algorithm is activated and a threat is detected, to other network-connected devices, may include, but is not limited to, a text message, phone call, email, push notification or other signal, visual or audible. In some embodiments, the system is programmed or set-up in such a way that end-users have the ability to let the system override a silent-mode setting on their network-connected mobile devices to receive an audible alert of a threat even if their mobile devices are on silent or do not disturb mode.

[0018] Preferably, the network-connected artificially intelligent object recognition algorithm is able to reliably identify a wide variety of objects and threats, and can identify weapons, faces, clothing and characteristics of individuals, and license plates and features of vehicles day or night, with a high degree of accuracy.

[0019] In some embodiments, there may be trained personnel, onsite or offsite, to assist the network-connected artificially intelligent object recognition algorithm in making a final determination to identify a weapon or other threat by reviewing real-time or historical footage or images flagged for further review by the artificially intelligent object recognition algorithm.

[0020] In some embodiments, the system is connected to a public safety answering point (PSAP), an emergency service dispatch center, or another government emergency service, so that trained personnel may be notified of threat detection through communication software integrating the PSAP, the emergency service dispatch center, or the other government emergency service with the system. For example, the communication software may connect to another system capable of analyzing and / or displaying camera streams and / or issuing notifications, such as a video management system or an emergency management system configured for the purpose and functionality of issuing alerts, that is hosted by the PSAP, the emergency service dispatch center, or the other government emergency service. In some embodiments, an international, federal, state or local public safety agency maintaining a crime database of individuals or vehicles of interest is connected to the system via the communication software.

[0021] In some embodiments, the characteristics of individuals being analyzed by the network- connected artificially intelligent object recognition algorithm may include the individual’s clothing and / or body shape, physique, height, gender, ethnicity or similar characteristics. These characteristics may be analyzed with or without the other characteristics mentioned.

[0022] In some embodiments, the network-connected artificially intelligent object recognition algorithm contains algorithms that are able to recognize unattended bags, oversized bags, gun bags, flags, suspicious clothing that could conceal a dangerous item, a fire, a gas leak, or other threats of the like.

[0023] In some embodiments, the user or other authorized person is able to set a threshold or a confidence level range for which the object detection alerts are automatically issued by the system, bypassing a manual review and immediately sending alerts to the other network-connected mobile devices , and another confidence level range for which those object detections alerts are sent inreal-time for review by a human person or persons, who makes the final determination if the alert should be sent to the end user.

[0024] In some embodiments, cameras in the system are continuously scanning for individuals or vehicles that are or could be threats, and create a database of individuals or vehicles that have been seen onsite within a given period of time in the past, and the system is able to automatically issue a real-time alert when an individual or vehicle that has not been seen onsite within a given period of time in the past is detected. The system can automatically activate, or users can manually activate from their mobile device or other application, a set of normally ‘dormant,’ but able to be activated, On-Demand threat recognition cameras.

[0025] In some embodiments, network-connected artificially intelligent object recognition algorithms are storing the footage from the streams for some or all of the network cameras either on the offsite and / or onsite network-connected server(s) or on a network video recorder (NVR) or equivalent device. The system is capable of searching and recognizing threats in the footage (as well as in the real-time streams cameras). In some embodiments, the user is able to quickly activate Al object recognition algorithms for threat recognition in stored footage corresponding to a given time period in the past, ideally at least one hour or less on a given set or cluster of cameras. For example, the user may be able to do this activation with a single button within the application on the mobile device.

[0026] In some embodiments, when onsite or offsite graphics processing unit(s) hosting the always active Al object recognition algorithms recognizing threats from the streams or frames of the cameras are experiencing reduced, below peak activity, they may be repurposed to process and analyze stored footage. Additionally or alternatively, the graphics processing unit(s) that are experiencing reduced, below peak activity, they may be repurposed to process and analyze realtime frames or streams from onsite cameras that are would otherwise be fed to a currently ‘dormant’ Al object recognition algorithms.

[0027] In some embodiments, the graphics processing units running the always active Al object recognition algorithms are located at the proximity of the cameras that feed the frames or streams from which the threats are recognized. For example, the always active Al object recognitionalgorithm run on a processor located within the camera housing or housed very near to it, generally within a few feet.

[0028] In some embodiments, there is a self-service portal, webpage or application for individuals seeking to set up their network-connected cameras with an On-Demand threat recognition system. In this scenario, there are very clear instructions for the integration of one’s cameras that require little to no third-party assistance to set up.

[0029] In some embodiments, the system is configurable so that a user or authorized administrator may preset for rapid activation certain cameras in preset clusters or zones that are configured to feed normally ‘dormant,’ but capable of On-Demand activation, threat detection algorithms. The cameras in these clusters or zones may be manually activated by a user on a network-connected device or other connected application, or may be automatically activated based upon a threat detection of a sufficient confidence threshold within a certain proximity range to the cluster or zones of cameras. A camera, or cluster or zone of cameras, configured to feed normally ‘dormant,’ but capable of On-Demand activation, threat detection algorithms, may be activated by mapping the coordinates of the cameras or clusters or zones of cameras and then setting a range, distance or radius from a threat detection of a sufficient confidence threshold to automatically activate them. Preferably, the preset clusters or zones that are capable of manual On-Demand activation (e.g., not based on range, distance or radius) include at least 10 cameras. In some embodiments, there is a single button within the application to select a particular preset cluster or zone that is configured to feed normally ‘dormant,’ but capable of On-Demand activation, threat detection algorithms.

[0030] In some embodiments, the cameras, or groups cameras in clusters or zones, which are configured to feed normally ‘dormant,’ but capable of On-Demand activation, threat detection algorithms, are able to be activated in increments of one second, one minute, one hour, and / or one day and those activations may be billed in a mostly proportionate way to the time of that the cameras, clusters or zones were activated.

[0031] In some embodiments, a photo, or a frame, with a person or vehicle of interest can be uploaded to the graphics processing units or can be selected from a stream, and the system is able to monitor in real-time where that person or vehicle of interest is across all or a subset of thecameras that are configured to feed normally ‘dormant,’ but capable of On-Demand activation, threat detection algorithms. The algorithms used to identify or track these persons or vehicles of interest may be trained to recognize faces, clothing, physique, vehicle features or type, and / or license plates, etc. In some embodiments, the user may be able to efficiently select a time period, such as 10 minutes, 1 hour, 1 day, etc., and cameras, clusters or zones of cameras for which the system will perform an efficient forensic search for threats matching the uploaded photo or selected frame. In some embodiments, by entering characteristics as text, such as, but not limited to “a person wearing a blue hoodie and black pants”, “a red sedan”, etc., a user is able to search efficiently for a threat a posteriori in not-previously analyzed, stored footage and / or streams from the cameras that are configured to feed normally ‘dormant,’ but capable of On-Demand activation, threat detection algorithms.

[0032] In some embodiments, all cameras in the system are configured to feed normally ‘dormant,’ but capable of On-Demand activation, threat detection algorithms, and are able to be activated by a network -connected device or other application.

[0033] In some embodiments, the set of cameras that are configured to feed always active threat detection algorithms, and the set of cameras that are configured to feed normally ‘dormant,’ but capable of On-Demand activation, threat detection algorithms are able to be changed, adjusted or rotated by an authorized user.

[0034] In some embodiments, the system may be integrated with an access control system and / or an emergency management system that can control locks, doors, etc. in a space. The system may have preprogrammed “traps” that will allow someone that has been identified as a threat, either by the system automatically or by an authorized user, and if that threat enters a given area, such as a stairwell or other enclosed space, and perhaps if there is no other person in that given area, the individual is able to enter that space but unable to exit the given area until an authorized user allows them to. In some embodiments, additional cameras that are configured to feed normally ‘dormant,’ but capable of On-Demand activation, threat detection algorithms, may simultaneously be activated when access control measures are taken.

[0035] In some embodiments, the system is integrated with a drone or drone system that is able to, autonomously or with human assistance, aid in tracking, disrupting, or eliminating a threat. In some embodiments, the threat is a drone or drones being identified and / or tracked by the system.DESCRIPTION OF THE DRAWINGS

[0036] The FIG. is a schematic view of an example of a system for detecting threats and issuing real-time alerts.DETAILED DESCRIPTION

[0037] The processes and systems described in the FIG. can improve the effectiveness of surveillance, security, or law enforcement response to threats by enabling the activation, via network-connected device, of normally ‘dormant’ artificially intelligent object recognition algorithms trained to detect threats in streams, images or frames from onsite network-connected security cameras, the normally ‘dormant’ artificially intelligent object recognition algorithms running on an offsite network-connected graphics processing unit or units. Any of the systems or components integrated into the greater system, may be directly integrated with any of the other systems or components in the greater system. Pareto’s principle roughly applies here, and it is true that for most buildings 80% of threats from weapons, individuals, or vehicles will be initially spotted on just 20% of network-connected security cameras that see the highest traffic or are on the perimeter of the facilities. By having these 20% of cameras always actively looking for threats, but the 80% of other cameras ready to be activated for active threat detection, it can be understood that this setup provides roughly 80% of the security, for about 20% of the cost when compared to having active threat detection running on all cameras at all times. The upfront investment for graphics processing units or the ongoing cost of renting server space from a cloud processing service is roughly 80% less in this situation, and there is only a small additional cost to rent the server space for the 80% of cameras during emergencies because you only have to pay a small fee when the cloud services are actually processing your data, and you wouldn’t have these 80% of normally ‘dormant’ cameras (i.e., cameras in the system are configured to feed normally ‘dormant,’ but capable of On-Demand activation, threat detection algorithms) actively searchingfor threats the vast majority of the time, only when a threat is suspected or detected or an incident is already underway.

[0038] In one example, the systems disclosed herein involve: an application running on and / or integrated with a network-connected device that can activate normally ‘dormant’ artificially intelligent object recognition algorithm(s) running on an offsite network-connected graphics processing unit or units, wherein the object recognition algorithm(s) is(are) trained to detect threats in streams, images, or frames, said artificially intelligent object recognition algorithm(s) running on an offsite network- connected graphics processing unit or units, onsite network-connected cameras are provided with a protocol or interface, preferably integrated, that can be used by offsite processing unit or units to receive (e.g., fetch) streams, images or frames to processing unit or units, the systems being capable of issuing alerts, audible or otherwise, to other specified, onsite or offsite, network-connected mobile devices or other electronic devices capable of receiving notifications.

[0039] In particular, the threats that the algorithms are able to detect may include, but are not limited to weapons, faces and characteristics of individuals, and features and license plates of vehicles.

[0040] In one example, the processes disclosed herein involve the following flow chart:1. A threat that is a weapon, specific individual, specific vehicle, or otherwise, is detected either by a user with a network-connected device, mobile or otherwise, that may have a specific software application, running on and / or integrated with it, or by a network- connected camera and processing unit, which may be processing onsite or offsite, that are actively detecting threats.2. Then, manually, at the command of the user with the network-connected device, mobile or otherwise, or, automatically, at the command of the network-connected processing unit actively detecting threats via onsite network-connected cameras or at the command of another software system that is associated and / or integrated with that network, an activation for threat detection occurs for a cluster or set of, as previously determined by a user or developer, network-connected 'dormant' cameras that were previously not actively detecting threats. This command or commands instructs a connection or activate an integration between this set of previously 'dormant' cameras and an offsite network- connected processing unit. This connection or integration may be achieved by using the RTSP or ONVIF streaming links, a VPN tunnel, a Video Management System (VMS) or some other means of direct or indirect integration. These 'dormant' cameras may be already in use in the sense that they are performing activities like recording streams, frames or images for processing and analyzing that data for marketing (like vehicle or body counting), quality assurance-related information (like process management), or other nonthreat related purposes, but these data are not currently processed or analyzed for threats, such as a weapon, specific individuals, specific vehicles, or otherwise.3. Once this connection or integration is activated the streams, frames, or images from the set of network-connected previously 'dormant' cameras that were previously not detecting threats are then, because of the command from the mobile device or always active camera and processing unit or other network-connected device, transmitted to the offsite network-connected processing unit. This offsite network-connected processing unit may be a specific processing unit in a specific location, as previously determined by a user or developer, or the streams, frames, or images may be sent for processing to a cloud computing service like AWS, Microsoft Azure, Google Cloud or similar service. There may be a notification sent to a set of previously determined, by the user or developer, network-connected devices, mobile or otherwise, and / or to an application containing information that the set of previously 'dormant' cameras has been connected and activated for active threat detection. Whether it is the streams, frames or images being analyzed and with what frequency they are analyzed for the previously 'dormant' cameras is something that is determined by the user or developer.4. This offsite network-connected processing unit then begins receiving the streams, frames, or images coming from the set of network-connected previously 'dormant' cameras and analyzes that data using an algorithm(s) trained to detect a set of, as previously determined by a user or developer, threats which may be weapons, specific individuals, specific vehicles or otherwise. The system is aware of which camera or location a stream, frame or image is coming in from because of the RTSP or ONVIF streaming link or other specific information, like a unique camera identification number, name or code, to identify the exact source of the images coming in with the frames.5. When a threat or threats is detected on the streams, frames, or images coming from the set of network-connected previously 'dormant' cameras by the threat detection algorithm(s) running on the offsite processing unit with a previously specified level of confidence, the system will then issue an alert or alerts to a set of, as specified by a user or developer, specific individuals, or devices, mobile or otherwise. Which individuals, devices receive notification of an alert or alerts is based on the identification information associated with the camera from which the threat was detected on that camera's stream(s), frame(s) or image(s). Furthermore, the contents of the alert, like what type of threat was detected at what confidence level, is determined by the threat detection algorithm running on the offsite processing unit and this type of information may also be included in the alert.6. The set of network-connected previously 'dormant', but now active, cameras that are having their streams, frames, or images analyzed by the threat detection algorithm on the offsite processing unit, may have the active threat detection deactivated for this same set of cameras by an authorized user or developer with a network-connected device, mobile or otherwise or the user or developer may set a specific amount of time before the previously 'dormant' cameras automatically become 'dormant' again.

[0041] These processes and systems disclosed herein may provide roughly 80% of the security for roughly 20% of the cost when compared to the best available options currently in existence, which is to have the data from many or all of your cameras at all times being evaluated for real-time threats.

[0042] Notably, in the United States, there is the national crime information center (NCIC) which provides a list of license plates associated with stolen vehicles or active warrants. The crime database may be connected to the system via integration with the communication software or manually uploaded to the system on a regular basis. Thus, an onsite camera can have its streams or frames processed by threat recognition algorithms running on an onsite (or offsite) graphics processing unit(s) for analyzing the license plates of vehicles in view of the camera. License plate data can then be compared by the system against the NCIC crime database or other list of vehicles of interest, and the system can issue alerts of any positive identifications of vehicles listed in the database. For example, the alerts can be sent to users and / or to the PSAP or other emergency services dispatch center. At this point, the system may automatically activate, or a user may manually activate previously ‘dormant’ Al object recognition algorithms running on the offsite network-connected graphics processing unit(s) so that onsite cameras can then have their streams or frames processed for threats, vehicles, individuals and / or weapons by the algorithms running on the graphics processing unit(s).

[0043] Furthermore, threat recognition algorithms running on an onsite (or offsite) graphics processing unit(s) can actively analyze faces of individuals in view of an onsite camera . Information on individual faces can be compared against a database of individuals of interest connected to the system. If there is a positive identification of individuals listed in the database, the system may automatically activate, or a user may manually activate previously ‘dormant’ Al object recognition algorithms running on the offsite network-connected graphics processing unit(s) so that onsite cameras can then have their streams or frames processed for threats, vehicles, individuals and / or weapons by the algorithms computer vision algorithms on an offsite graphics processing unit(s).

[0044] Specific embodiments of the invention have been disclosed. It should be understood, however, that the detailed description is not intended to limit the invention to the particular form disclosed, but on the contrary, the intention is to cover all embodiments delimited by the claims and their equivalents.

Claims

CLAIMS1. A system for On-Demand threat detection on a site and real-time alerts, comprising: a network communication infrastructure; a first network-connected offsite graphics processing unit(s); a first artificially intelligent (Al) object recognition algorithm trained to detect threats in streams, images or frames, wherein the first Al object recognition algorithm runs on the first network-connected offsite graphics processing unit(s), wherein the first Al object recognition algorithm is normally ‘dormant,’ and wherein the first Al object recognition algorithm is configured to be activated when the first Al object recognition algorithm is dormant; a first network-connected cluster or set of security cameras located onsite and configured to feed the first Al object recognition algorithm with the streams, images or frames; and means for activating the first Al object recognition algorithm.

2. The system of claim 1 wherein the means for activating the Al object recognition algorithm comprises on or more of: a network-connected device running on / and or integrated with an application programmed to activate the Al object recognition algorithm upon a user command; or a second network-connected graphics processing unit(s), a second artificially intelligent (Al) object recognition algorithm trained to detect threats in streams, images or frames, wherein the second Al object recognition algorithm runs on the second network-connected graphics processing unit(s), wherein the second Al object recognition algorithm is active, and a second network-connected security camera(s) configured to feed the second Al object recognition algorithm; ora network-connected microphone or audio sensor, and a processing unit that has an algorithm trained to identify a gunshot.

3. The system of claim 2 wherein the means for activating the Al object recognition algorithm comprises the network-connected device, and wherein the network-connected device is mobile or fixed.

4. The system of claim 2 wherein the means for activating the Al object recognition algorithm comprises the second network-connected graphics processing unit(s), and wherein the second network-connected graphics processing unit(s) is located onsite or offsite.

5. The system of claim 4 wherein the second network-connected graphics processing unit(s) is located within or near a housing of the first network-connected security camera.

6. The system of claim 4 wherein the second network-connected graphics processing unit is programmed to process and analyze real-time frames or streams from first network-connected security camera when the second network-connected graphics processing unit(s) experiences reduced, below peak activity.

7. The system for claim 2 wherein the second network-connected graphics processing unit reports Al object recognition algorithm results of a vehicle or person (likely license plate or a face) that matches a vehicle or person in a crime data base, and the first artificially intelligent (Al) object recognition algorithm is activated for the first network-connected cluster or set of security cameras and / or that positive match is communicated to an emergency management center through a direct or indirect integration.

8. The system of claim 1 wherein the first network-connected graphics processing unit(s) is located offsite.

9. A method for On-Demand threat detection on a site and real-time alerts, comprising: providing: a network communication infrastructure; a first network-connected offsite graphics processing unit(s); a first artificially intelligent (Al) object recognition algorithm trained to detect threats in streams, images or frames, wherein the first Al object recognition algorithm runs on the first network-connected offsite graphics processing unit(s), wherein the first Al object recognition algorithm is normally ‘dormant,’ and wherein the first Al object recognition algorithm is configured to be activated when the first Al object recognition algorithm is dormant; a first network-connected cluster or set of security cameras located onsite and configured to feed the first Al object recognition algorithm with the streams, images or frames; and means for activating the first Al object recognition algorithm; and activating the first Al object recognition algorithm.

10. The method of claim 8 wherein the means for activating the Al object recognition algorithm comprises on or more of: a network-connected device running and / or integrated with an application programmed to activate the Al object recognition algorithm upon a user command; ora second network-connected graphics processing unit(s), a second artificially intelligent (Al) object recognition algorithm trained to detect threats in streams, images or frames, wherein the second Al object recognition algorithm runs on the second network-connected graphics processing unit(s), wherein the second Al object recognition algorithm is active, and a second network-connected security camera configured to feed the second Al object recognition algorithm; or a network-connected microphone or audio sensor, and a processing unit that has an algorithm trained to identify a gunshot.

11. The method of claim 9 wherein the means for activating the Al object recognition algorithm comprises the network-connected device, and wherein the network-connected device is mobile or fixed.

12. The method of claim 9 wherein the means for activating the Al object recognition algorithm comprises the second network-connected graphics processing unit(s), and wherein the second network-connected graphics processing unit(s) is located onsite or offsite.

13. The method of claim 11 wherein the second network-connected graphics processing unit is located within or near a housing of the first network-connected security camera.

14. The method of claim 11 wherein the second network-connected graphics processing unit is programmed to process and analyze real-time frames or streams from first network-connected security camera when the second network-connected graphics processing unit experiences reduced, below peak activity.

15. The system for claim 2 wherein the second network-connected graphics processing unit reports Al object recognition algorithm results of a vehicle or person (likely license plate or a face) that matches a vehicle or person in a crime data base, and the first artificially intelligent (Al) object recognition algorithm is activated for the first network-connected cluster or set of security cameras and / or that positive match is communicated to an emergency management center through a direct or indirect integration.

16. The method of claim 1 wherein the first network-connected graphics processing unit is located offsite.

Images