Real-Time Face Tracking Based on the Kalman Filter

Abstract

Face tracking is not only one of the most interesting tasks in the fields of computer vision, but also an important part of practical applications such as surveillance systems, fault face detection systems, artificial intelligence, etc. Face tracking systems can be difficult to track the human face when face detection has too low of an accuracy rate. This paper proposes a method for real-time face tracking based on the Kalman filter. The proposed method has two main parts, such as face detection by using the YOLOv3 (You Only Look Once) algorithm and face tracking by using the Kalman filter. The face detection system is difficult to detect a human face when the face is occluded or other problems are affected (poor lighting conditions, motion blur, etc.), then the Kalman filter predicts the face's future location. The experimental results show that the proposed method can track the human face not only in different face position but also at high speed in real-time. Testing for this work used the Image Processing Toolbox and the Deep Learning Toolbox in MATLAB application 2021a.