Human Eye Tracking Based on CNN and Kalman Filtering

Abstract

The driver fatigue detection method based on human eye feature information has the advantages, such as non-invasion, low cost, natural interaction and so on, which has been widely favored. However, in the actual detection process, the driver’s face will be shaken due to various factors, and there will be motion blur, which will cause misjudgment and missed judgment on the fatigue driving detection. Therefore, this paper designs a method based on CNN convolutional neural network to detect human key points, then uses Kalman filter to track human eyes, eliminates jitter interference, and greatly improves the accuracy of fatigue detection. The experimental results show that the proposed method can track the human eyes in real time and has high accuracy and robustness.