Fusing Frontal Face Recognition Using Multi View Cameras

Abstract

The recognition of faces in videos has recently gained considerable attention, but the recognition process executed on a single camera has limitations, especially when dealing with uncooperative subjects, changes in body posture, or self-occlusion. These challenges are particularly relevant in the context of studying facial analysis in children with Autism Spectrum Disorder (ASD). Therefore, the use of multiple cameras in a face recognition system is proposed to overcome these limitations. Facial image realignment was employed in the automatic face recognition process. To achieve this, the Kanade-Lucas-Tomasi (KLT) algorithm was used to track facial features, and the RANSAC algorithm was utilized to estimate the homography transformation for realigning the multi-view input images. To assess and compare the similarity of the fused image, the normalized cross-correlation (NCC) was employed. The resulting fused image was obtained based on the extracted pose of the face. The results demonstrate the efficacy of the method, achieving an accuracy of 94.5% for typically developed children and 87.3% for ASD children.