A team of Asian researchers said they have developed a vision sensor that closely emulates the way human retinas adapt to light, which they believe will improve machine vision for facial recognition.
The team comes from Hong Kong Polytechnic University and Yonsei University in South Korea and is led by Chai Yang, an associate professor in PolyU’s Department of Applied Physics.
Together they worked to improve the adaptation of lighting in machine vision systems that use circuits and algorithms. This adaptability would allow artificial vision to recognize objects more effectively in complex environments.
The researchers developed bio-inspired systems that use light detectors, known as phototransistors, using a double layer of ultra-thin molybdenum disulfide, a semiconductor with unique electrical and optical properties. There the team added ‘charge trap states’, which are impurities or imperfections in the crystal structure of a solid that restrict the movement of a charge, to the double layer.
Trap states can store light information and dynamically modulate the optoelectronic properties of a device at the pixel level. In effect, the trap indicates the control of the movement of electrons to adjust the amount of electricity conducted by the phototransistors. This allowed scientists to control the device’s ability to detect light, its photosensitivity.
Just as the human eye is made up of rod and cone cells that detect dim light and bright light respectively, newer vision sensors have arrays of phototransistors. In their paper, the team writes that the vision sensor can distinguish objects in less than optimal lighting conditions and can switch between and adapt to different light levels.
Facial biometrics remain relatively dependent on good lighting conditions for high precision matching.
The experimental sensor would surpass not only current sensors, but even the human eye. The human eye has a range of around 160 decibels (dB), while conventional silicon-based sensors reach 70 dB. The vision sensor developed by the researchers is said to have an effective range of up to 199 dB, exceeding both.
“The sensors reduce hardware complexity and dramatically increase image contrast under different lighting conditions, providing high image recognition efficiency,” Chai says.
According to the research paper, potential applications for the sensors are facial recognition, autonomous vehicles, the Internet of Things, and edge computing environments.
The research, titled, ‘Bioinspired visual adaptation in the sensor for accurate perception,’ was published in the journal Nature Electronics.
Another recent development in biometric and light capture comes from IDloop. A company researcher demonstrated the use of structured light rather than conventional scanner architecture. The structured light is said to provide superior non-contact 3D fingerprint capture with improved registration and less distortion.
precision | biometrics | facial recognition | image recognition | machine vision | research and development | vision sensor