SMART LOW-POWER EYE TRACKING WITH ON-LENS TRANSPARENT PHOTODETECTORS AND EMBEDDED PROCESSING
The project aim is the development of a hardware demonstrator for low-power eye tracking with lens-integrable transparent photodetectors and embedded machine learning. In parallel to this main topic, two secondary studies are carried out. The first one regarding the integration of an energy harvesting system based, for instance, on flexible organic photovoltaic materials (whose integration in the glass frame is far easier and aesthetically appealing than inorganic photovoltaics). The second study, instead, regards the scenario recognition, i.e. the ability of distnguishing the indoor environments (such as offices, houses, and so on) from the outdoor environments.
An important research activity will be the evaluation of the integrability of the developed components (photodetectors and electronics) with the smart glasses. The photosensors should be embedded within the lens and the rim, LEDs in the frame, while the electronics and batteries within the temples. A prototype integrating all the components will be fabricated. Assembly tests of the parts realized with 3D printing and resin molding techniques will be carried out.
The demonstrator will include invisible (NIR) light sources (LEDs) with their driving blocks, an array of photodetectors collecting the light from eyes, an analog front-end for the readout of the photodetectors signals, a local data acquisition and processing system based on low-power microcontroller for reconstructing the direction of gaze, radio peripherals (such as Bluetooth Low Energy modules) allowing for wireless cummunication with other devices, and a power management system managing the smart glasses battery. The work builds on two activities. The first was a feasibility study of the use of organic photodetectors which, being transparent and flexible, can be placed and/or directly developed onto the lens, thus dramatically simplifying the reconstruction of the direction of gaze. The second was the realization of a preliminary electronic demonstrator including all the elements stated above. Miniaturization and low power consumption will be the key guidelines for the design of the electronics.
In parallel to the development of the electronic hardware, the most suitable processing algorithms for real-time, embedded, robust and accurate reconstruction of the direction of gaze (and potentially other ocular parameters) will be investigated. In particular, the use and the efficiency of embedded machine learning will be studied. Analog vs. digital implementation of the processing will be compared both in simulation and experimentally, verifying the feasibility of developing a single microelectronic circuit (ASIC) dedicated to this task. This ASIC would include the sensors front-end circuit, as well as a neural network of medium complexity to process the signals at the Edge, and it would directly provide to the MCU the direction of gaze.
Back to Alumni