In the convergence between IoT technologies and home security systems, the result has been the emergence of a new class of smarter intruder alarms—mechanisms boasting improved accuracy in detecting intrusion into the home while reducing the chance of false alarms.
At the heart of the automation and enhanced functionality is a process involving machine learning—a popular concept that is so often interchanged, however, with deep learning and artificial intelligence (AI). As a result, the subtle distinction among the three concepts is all but lost, leading to confusion or even overblown expectations as to what these processes can do, in this case, to boost the effectiveness of security systems.
What is—and isn't—AI
Artificial intelligence is an umbrella term for various data analytics processes. As an example, a computer imbued with artificial intelligence will process all inputs through its so-called neural networks and via other statistical methods. Constantly evolving, these networks generate outputs that are adjusted based on variances in the data being fed to the system. The computer then figures out, independently, the appropriate responses to make in relation to the input data—without the need for the user's input.
Machine learning is a subset of AI, and refers to a process that uses statistical techniques to give computers the ability to "learn" with data, without explicitly programming the computer to do so.
Deep learning is a subset, in turn, of machine learning, denoting processes based on deciphering the significance and meaning to be derived, if any, from the input data. The opposite of the deep learning process would be task-specific algorithms that result in causing either action A or B. With deep learning, however, the resulting output from the system will have considered specific features of the input data, paving the way for action that is more accurate as well as proportional to the received data. For instance, if a security camera sees a person walking or behaving differently from others, the system may deem the object to be a threat and then goes on to activate the alarm system within the premises.
The beginnings of AI in home security systems
There are two applications where artificial intelligence is used currently in home-security systems.
The first is in systems integrated with consumer video cameras. Here video cameras use facial recognition functionality to identify whether the moving object is an intruder or a home member. Facial recognition in such a setting represents a basic form of machine learning, where a system analyses detected objects in the video footage against a pre-determined set of previously uploaded and approved images. Based on the results of the analytical process, the system sounds an alarm if the object doesn't match any of the images.
The second application in intruder alarm installations is when the system is integrated with a voice assistant. In this case, skills in the voice assistant have been developed with the use of numerous artificial intelligence methodologies to ensure smooth interaction between the user and the device.
Automated doesn't mean it's AI
Some current functionalities of intruder alarm systems may be confused with artificial intelligence, especially when the former play an important role in automating interactions with both the system and any system-connected smart home devices.
The automatic arming and disarming of a system is a key example. There are two ways this can be set up on an intruder alarm system.
The first involves specifying the time window within which the intruder alarm system will automatically arm, while also stipulating the equivalent time window when the system disarms. This can be done by the installer on behalf of the end-user through a keypad. Alternatively, the end-user can set time windows through an attached mobile phone application. Strictly speaking, what has been described isn't an artificial intelligence process because the time windows are arbitrarily set by the end user or installer, and any changes will require interaction from the user. In contrast, deploying AI to automate the arming and disarming action means no user input is necessary to adjustor set the time windows.
Another way of automating the arming/disarming process is through geofencing. In intruder alarm systems, geofencing uses the current GPS location of the user's phone to identify whether it is present in a designated "detection" area—usually the perimeter around the user's house extended out for a couple of hundred meters. Once the user's phone is sensed as having entered the detection area, the intruder alarm system disarms. If home automation devices are integrated with the intruder alarm system, geofencing can also open the garage door, unlock the front door, and/or turn on the lights in the yard or inside the property.
Potential applications of AI in home security systems are myriad
Although current applications of AI processes in intruder alarm systems remain limited, many potential applications are possible that could give home security systems a much further boost, delivering even higher levels of security and comfort in the house.
In the future, artificial intelligence can lead to the creation of a true smart home capable of learning the ways, habits, and preferences of the occupants of a home, automatically adjusting the settings in its various systems to accommodate both normal patterns and quirks in behaviour.
As more and more intruder alarm systems are integrated with home automation devices, genuine AI capabilities—where the system performs actions independent of the user's input—could create a truly smart and secure home in which security and comfort at home are greatly improved. In contrast, current home automation devices still rely on pre-determined scenarios that the user or installer must set up, following very specific criteria. Any changes to the settings must also be done manually by the system operator.
It could also actively monitor activity in the house and arm the systems at times when home members are out in the garden while the property remains unlocked and unoccupied for prolonged periods of time. If in such a circumstance a child were to run back into the house while the system is armed, the cameras could quickly recognise that it is not an intruder and not sound the alarm.
AI processes could be used by alarm receiving centres to create databases of all incoming alerts, which then could be analysed to predict false alarms. Initial pre-assessment of an incoming alert by the AI software could help the operator respond faster to alerts that are least likely to be false. At the same time, the software and system will be able to detect with near-accurate certainty the shape and form of a false alarm, helping to reduce costs and resources otherwise diverted to deal with false alarms.
Intrusion systems integrated with cameras can also help protect children against kidnappings. Cameras monitoring the area occupied by a child could sound an alarm upon the approach of the kidnapper, while also recording the kidnapper's image and sending it directly to a monitoring centre or to police authorities.
Anna Sliwon is an Analyst for Security Technology within the IHS Technology Group at IHS Markit
Posted 8 August 2018