Support Vector Machine-Based Face Direction Detection Using an Infrared Array Sensor

Abstract

Facing direction detection plays a critical role in human computer interaction, such as face recognition and head pose estimation in biometric identification, spatial microphone/loudspeaker devices, virtual reality games and etc. Currently detection methods are mainly focused on extracting specific patterns of various facial features from the user’s optical images, which raises concerns on privacy invasion and these detection techniques do not usually work in the dark environment. To address these concerns, this paper proposes a pervasive solution for a coarse facing direction detection using a low pixel infrared thermopile array sensor. Support vector machine (SVM) method is selected as the classifier. Two methods for feature extraction are compared. One is based on pre-defined features and the other is based on pre-trained convolutional neural network (CNN) model. The detection accuracy resulted from using pre-defined features reaches 87% for identifying five different facing directions up to 1.2 m. However, this accuracy is largely descended when the detection range is increased to 1.8 m. Interestingly, the accuracy resulted from using pre-trained CNN features, however, demonstrates a reliable and steady performance up to 1.8 m. The accuracy keeps above 95% at both detection ranges (1.2 and 1.8 m). This proposed solution leads to many advantages, such as low resolution leading to no intention on privacy invasion, and the low-cost intriguing a potentially large market for human computer interaction in smart home appliances control and computer games.