A Face Tracking System under Various Illumination Environments
Date Issued
2005
Date
2005
Author(s)
Lin, Yuan-Pin
DOI
zh-TW
Abstract
The goal of Human-Computer-Interaction (HCI) is to recognize users’ intention through their gestures and speeches, so that users can executive certain tasks on PC without using traditional input devices. Therefore, how to improve the stability and precision of the recognition process is a crucial topic of HCI. In this study, we focus on the issue of face detection to ensure the recognition of users’ gestures based on robust face feature.
Skin-tone is luminance-dependent and highly sensitive to variations in background illumination. Thus, the performance of a face tracker based solely on the skin-tone would be degraded dramatically when the environment changes. Nowadays, with the prevalence of webcam device and notebook, the issue of illumination effects would arise while using portable devices under various environments.
In this study, we present an effective illumination recognition technique combining K-Nearest Neighbor classifier and an adaptive skin model. With this approach, our system can derive an optimal skin model for the face tracking task under various illumination environments. In the static simulation stage, users’ faces are captured via webcam under five typical illumination conditions in daily life, including 50 images for training procedure and another 100 images for testing task. Based on the simulation results, we have demonstrated that the recognition rate of illumination of KNN is 98% and the accuracy of skin detection is 90.90%. In the real-time implementation, the system successfully tracks users’ face at frame rate 15 fps under standard notebook platforms, and it only takes 45% of window resource. Furthermore, this system permits users to define and add their favorite environments to the KNN classifier to increase the flexibility of our face tracking system.
In the next stage, we will focus on one of HCI application – Webcam Mouse. Relative motion vectors between eyes and face features will be used to control the computer cursor via head rotation. Then, we would implement this system into portable devices, such as PDA and smart phone, to make this application more friendly and practical.
Subjects
人機互動
臉部追蹤
Human-Computer-Interaction
face detection
Type
thesis