High-speed gaze detection using a single FPGA for driver assistance systems

Abstract

The performance of driver gaze detection by video-based eye-tracking often encounters problems in lowcomputing speed, high-power consumption, and installation space constraints inside the vehicle. In this paper, we present an eye-tracking system that uses a single field-programmable-gate-array chip to overcome the aforementioned problems. In the detection system, the image quality is 640 $$\times$$ 480 pixels with an 80 fps frame rate. Eye feature extraction is conducted using the enhanced semantics-based vague image representation approach. A succinct fully-connected neural network is then employed to classify various directions of sightline. Our experimental results exhibited a noticeable recognition speed at 0.52 $$\upmu$$ s using a 100 MHz system clock and had an average detection rate of 92%.