Illinois and ADSC researchers teamed up with startup Inspirit IoT to develop a promising low-cost solution: an audio-based security system that can alert a central command center when trouble breaks out. The system, which measures about 10 centimeters in length and can be built for around $20, recently won first place at the IEEE/ACM Design Automation Conference (DAC) 2017 International Hardware Design Contest this summer.
Winning DAC-IoT system
“We’re taking a big step forward in terms of the quality of audio event analysis on a very inexpensive platform,” said ADSC researcher Deming Chen, a professor in the Coordinated Science Laboratory and of Illinois’ Electrical and Computer Engineering Department. “We’re no longer going to be limited to providing security systems for only the most high-value locations where it’s worth spending a lot of money. If we can do it cheap, we can put microphones all over the place for a variety of applications.”
This system includes four microphones and a field-programmable gate array (FPGA), a type of integrated circuit, which has been configured to detect and classify audio sounds. If the system recognizes a suspicious sound, such as gunfire or screaming, it could alert a human operator at a central command station and direct police to a general location using beamforming.
The technology would employ “smart” sensors that only report suspicious activity, which both protects privacy and preserves computational power, says ECE Lecturer Zuofu Cheng, who is also a part-time employee of Inspirit IoT and helped develop the winning design.
“Our device is designed in real-time to classify events as opposed to recording audio and streaming it somewhere else, which means no one is spying on everything people say,” Cheng said. “The system might detect certain key words and trigger notification for those key words, which means we’re processing a lot less data but more targeted data.”
Similar devices that detect gunshots are already deployed in large cities. But the group’s design also identifies human screaming, which is a much difficult sound to detect, says Di He, a PhD student in Chen’s lab.
“When detecting these trigger sounds is paired with event direction recognition, it resulted in a unique system that serves more practical use,” says He, one of four students who contributed to the work. “We also published a research paper in ‘interspeech’ introducing an audio feature to improve screaming detection performance at low computational cost.”
The technology could work as an audio-only system or it could be combined with a network of security cameras, which are effective in providing even more comprehensive coverage with reduced false alarms.
The startup company, Inspirit IoT, was co-founded by Chen and two research scientists Kyle Rupnow and Swathi Gurumani of the Advanced Digital Sciences Center, an Illinois research center in Singapore. The company is designing an IoT (internet of things) platform that leverages machine learning to help simplify deployment of computer vision and smart sound applications in automotive, sports and entertainment, consumer, robotics and machine vision, medical, and security/surveillance domains.
“This concept behind this audio system can be used for many IoT applications,” said Kyle Rupnow, chief technology officer of Inspirit IoT. “We’re working to provide end-to-end solutions in different market segments.”
CSL Director Klara Nahrstedt believes the system could play a significant role in broader smart community initiatives at the University.
“Sensing devices are critical as we work to build smarter, safer communities,” said Nahrstedt, the Ralph M. and Catherine V. Fisher Professor of Computer Science. “This product provides a low-cost, commercially viable solution to enable safety for smart communities in diverse environments while protecting privacy and promoting the greater good."