Fast Eye Detector Using CPU Based Lightweight Convolutional Neural Network
Muhamad Dwisnanto Putro, Duy-Linh Nguyen, Kang-Hyun Jo · 2020
Eye detection is a crucial task for the success of driver drowsiness detection. The eye location is determined by employing feature extractions to discriminate distinctive features. Convolutional Neural Networks (CNN) has achieved the best results in the object detection task. However, this requires expensive computation, whereas practical application demands for this work to run real-time on low-cost devices such as CPU. The eye feature is very different from other organ facial features, so the shallow architecture of CNN deserves to be employed. This paper proposes the Fast Eye-CPU (FE-CPU) as a real-time eye detector that can work on the CPU. The architecture consists of two main modules, including the backbone to rapidly extract features and the detection module for predicting eye regions. As a result, the detector achieves high accuracy on several benchmark datasets, and it can work in real-time by 467 frames per second on Intel I5-6600 as a CPU device.