How Security Devices “Listen” To Detect Physical Threats Real Time
Imagine the following scene. A 6' tall male college student in T-shirt and running shorts is walking down an empty street at night. He hears a gunshot near him. Terrified, he screams and starts running.
Nobody’s got hurt, but a college student has certainly encountered a physical threat.
How could CCTV view this scene as threat just from looking at the moving image? Wouldn’t it look like a boy jogging at night in running shorts?
Recognizing Human Limitation
When it comes to surveillance, such as identifying suspicious activities and crimes, video monitoring often plays a crucial function.
It is however, impossible for human operators to see and detect everything. There are many factors that deter operators from accurately spotting suspicious behavior: number of screens monitored, level of alertness, bandwidth issues on cameras, room environment, etc.
As a result, millions of hours of video footage is often used only for post-incident investigation, instead of catching crime when it is happening.
Research indicates that the vast majority of gunfire discharges go unreported. Police officers cannot effectively respond to incidents of which they are not aware.
Sound Event Detection: Detecting Threats Real-Time
That’s how sound event detection came into play. Sound event detection, a machine listening technique, aims to detect specific sound event that occurs in an audio recording or real time streaming. Sound events that might imply emergency include glass smashing, gunshot, screaming and more.
Let’s go back to the imaginary scene.
Following the gunshot and screaming in the street, a security operator is immediately notified of the event. In less than 60 seconds, the operator reports the event and the police is dispatched.
As shown, Sound event detection, supported by inexpensive acoustic sensors, can provide valuable situational awareness because the sound is less affected by obstacles or field of view.
With that, let’s look at other areas of physical security where sound event detection can reach its full potential.
Autonomous Security Robots Physically Deterring Crimes
With video cameras that act as eyes and microphones as ears, security robots have emerged as an effective solution for automating security operations.
According to the US Department of Justice, visible video surveillance only deters crime by up to 40% under the best of circumstances.
Robots provide visible 360-degree eye-level video surveillance. With its 400-pound roaming physical presence, it can serve as a powerful deterrent for unwanted behavior or individuals intending to commit crime.
Upon hearing car door slamming for instance, robots can locate the source of the sound, its direction and distance, approach the source and take action.
There are many repetitive tasks routinely performed by human security officers on a daily basis. If these processes can be automated, humans focus on the strategic decision making and enforcement work.
Indoor Security: Your Trusted Watchdog When You Are Away
There is a little smart device that one third of American population possess which can act as a home guard — the smart speaker.
Since sound event detection only requires a microphone, it can be running on most of smart device: smart phone, smart watches, speakers, door locks, etc.
If you own an Amazon Echo, Amazon’s smart speaker, you can protect your home with a handy feature called Alexa Guard. With this option enabled, Alexa can send you notifications if your Echo detects certain sounds, like smoke alarms or breaking glass.
With connected smart lights, Alexa Guard can also turn the lights on and off to make it seem as if someone is home. Smart Alerts can be sent to certain home security system providers like ADT.
Automobile: Secure your Car and Passengers
Breaking into cars by smashing windows happens frequently — so frequently that in San Francisco, district attorney is proposing reimbursing residents whose windows are smashed.
Most dash cams use G-sensor to detect unusual events like sudden acceleration and deceleration. However, it has limitations detecting events that happen while parked, such as smashing windows.
If you put your car into parking surveillance mode and enable sound event detection, it will be able to detect emergencies and warn you of suspicious events.
Sound event detection can also help to save unattended passengers. Several children are left alone on school buses. On hot days, few hours are enough for a child to die from heat stroke, not to mention dogs.
Cochl. Sense: The ultimate tool for environmental sound detection
Analyzing environmental sounds in a real-world used to be almost impossible because there are lots of variations in target sounds, recording devices, and environments.
World’s first and only AI-based machine listening API, Cochl. Sense was trained in noisy real world environment. Its accuracy and performance was recognized in IEEE DCASE, world’s biggest challenge for sound detection& classification.
It supports both Cloud API and Edge SDK which means that it can be deployed literally anywhere and business clients as well as individual developers can easily integrate our machine listening technology into their products and services, in less than 5 minutes of time.
