A Head Pose Estimation Method Based on Stacked Autoencoder

Abstract

The invention discloses a head posture estimation method based on stacked self-encoding, which belongs to the technical field of computer vision. The main idea is to propose the use of stacked autoencoders to establish a nonlinear mapping relationship between head depth images and poses. The present invention first collects a large number of head depth images as training samples, extracts gradient direction histogram features at the same time, and then records the corresponding head poses. After that, a stacked autoencoder is designed, and the parameters of each layer of the stacked autoencoder are learned by using the gradient descent method on the training samples and calibration pose data. Finally, for the head image whose pose is to be estimated, the gradient direction histogram feature is extracted, and the head pose is estimated according to the above learned stacked autoencoder. Compared with traditional head pose estimation methods, this method can simulate the complex mapping relationship from input features to head pose, effectively overcoming the problem of low estimation accuracy of shallow models.

Description

technical field [0001] The invention belongs to the technical field of computer vision and relates to the problem of head pose estimation in images. Background technique [0002] Head pose estimation (e.g. figure 1 ) refers to accurately and quickly estimating the deflection angle of the corresponding head in the image based on the digital image of the head by using machine learning and computer vision methods, also known as the head pose. It is a hot issue in the field of computer vision and machine learning in recent years, and has a wide range of applications in human-computer interaction, safe driving, and attention analysis. For example: in the field of human-computer interaction, the deflection angle of the head can be used to control the direction and position of the computer or machine display; in the field of safe driving, the head posture can be used to assist in the estimation of line of sight, thus prompting the driver to the correct line of sight direction. In...

AI Technical Summary

Problems solved by technology

Classifiers commonly used in this type of method include Support Vector Machine (Support Vector Machine, SVM), Linear Discriminative Analysis (LDA), and Kernel Linear Discriminative Analysis (KLDA). The main disadvantages of this type of method are It is impossible to estimate the output continuous head pose, see the literature: J.Huang, X.Shao, and H.Wechsler, Face Pose Discrimination using Support Vector Machines (SVM), International Conference on Pattern Recognition, pp.154-156, 1998

Images