According to a recent report in Applied Physics Letters, Oregon State University in the United States has made significant progress in the research and development of new optical sensors. They developed a new type of optical sensor that can more realistically mimic the human eye's ability to perceive changes in objects. This achievement is expected to bring major breakthroughs in the fields of image recognition, robotics, and artificial intelligence.

The current information processing algorithms and architecture have become more and more like the human brain, but the way of receiving information is still designed for traditional computers. In order to realize its full potential, a computer that "thinks" more like the human brain needs an image sensor that "sees" more like the human eye.

The human eye contains about 100 million photoreceptors, but there are only one million connections between the optic nerve and the brain, so a lot of preprocessing and dynamic compression must be performed in the retina before images are transmitted.

Traditional sensing technologies, such as chips in digital cameras and smart phones, are more suitable for sequential processing. Each sensor generates a signal whose amplitude varies with the light intensity it receives, which means that a static image will cause the sensor to produce a relatively constant output voltage.

In the new retinal morphology sensor, the unique photoelectric properties of perovskite are used. Perovskite is placed in an ultra-thin layer with a thickness of only a few hundred nanometers to act as a capacitor. Under light, it changes from an electrical insulator to a conductor. Therefore, the sensor remains relatively quiet under static conditions. When it detects a change in light, it will record a short and sharp signal, and then quickly return to its baseline state.

Researchers simulate a series of retinal morphology sensors to predict how the retinal morphology camera responds to input stimuli. For example, in a simulation demonstration of baseball practice, infield players are displayed as clearly visible and bright moving objects, while relatively stationary objects such as stands gradually disappear. Even more striking is that a bird flew into the field of view, then stopped on an invisible bird feeder, almost disappeared, but reappeared during takeoff.

Researchers can also input any video into these simulations and process the information in basically the same way as the human eye. For example, let the robot use these sensors to track the movement of the target, and any static state in its field of view will not cause a response. Once the target moves, high pressure will be generated, and the robot will be told the position of the target immediately without any complicated image processing .

This new type of sensor can also be perfectly matched to neuromorphic computers. Neuromorphic computers are different from traditional computers in that they are a massively parallel network that simulates the human brain, providing support for the next generation of artificial intelligence used in autonomous vehicles, robotics, and advanced image recognition.

More than 80% of the information processed by the human brain is obtained through the eyes, and the information processing ability of the visual system is largely dependent on the structure and function of the retina. Therefore, it has always been the dream of many engineers to build a retinal sensor that can be comparable to the human eye and can simultaneously perform information detection and processing functions—or, in other words, an eye that truly mimics natural creatures. In recent years, very valuable breakthroughs have begun to appear in this field. For example, the optical sensor in this article is one of them. However, it may take at least many years to see this technology go into practical application.