Existing frontal pose estimation tasks rely on the estimation of the 3D pose of the head and the use of empirical thres-holds.This type of method has problems of subjectivity and insufficient robustness.To solve the above problems,an ensemble learning algorithm driven by large-scale frontal face data was proposed.The problem of large intra-class variance and small inter-class variance in frontal face posture classification was solved by constructing a large-scale frontal face category,and the subjec-tivity problem caused by artificially determining thresholds was avoided.The posture information in facial features and large-scale integration was used to distinguish frontal and non-frontal images,classification capabilities were improved,and the robustness was enhanced.Experimental results show that the proposed method does not need to rely on key point annotation,has short inference time,and achieves frontal pose estimation on public data sets.Efficient classification capabilities are demonstrated on real data sets with illumination changes,accessory occlusion,small angles and large angles.
维普
万方数据
