Font Size: a A A

Research On Head Pose Estimation Method Based On Deep Learning

Posted on:2022-08-21Degree:MasterType:Thesis
Country:ChinaCandidate:H H DaiFull Text:PDF
GTID:2518306551470974Subject:Computer technology
Abstract/Summary:PDF Full Text Request
In recent years,with the improvement of computer performance and the rapid development of deep learning,computer vision has become a hot research field.As a branch of computer vision based on biological characteristics,head pose estimation is a research direction of many scholars.The definition of head pose estimation is the process of inferring the head orientation of a person in a three-dimensional space from a two-dimensional portrait image.The research of head pose estimation has a wide range of application scenarios in many fields,such as driver monitoring system,virtual reality,security monitoring system,student classroom attention estimation.At present,head pose estimation still faces many challenges,such as low accuracy and weak model generalization ability.Based on deep learning,this paper improves the head pose estimation method from many aspects and proposes three head pose estimation algorithms.These methods have effectively improved the prediction accuracy and generalization ability of the model.The main research tasks of this paper include:(1)This paper proposes an anti-aliasing head pose estimation model,which solves the aliasing phenomenon that occurs in the downsampling process of traditional convolutional neural networks.This model adds anti-aliasing convolution operation to simulate low-pass filter operation in the head pose estimation network,which effectively alleviates the aliasing phenomenon that appears in the feature extraction network,and enables the network to extract more accurate features.Experiments on multiple public dataset demonstrate that compared the ordinary model,the anti-aliasing head pose estimation model has higher prediction accuracy.(2)Based on the anti-aliasing feature extraction network,this paper proposes a multi-task head pose estimation model.The model uses facial landmarks detection as an auxiliary task of head pose estimation.The algorithm explores the potential connection between facial landmarks detection and head pose estimation by optimizing two loss functions at the same time,and then efficiently improves the accuracy of head pose estimation accuracy.(3)Based on the anti-aliasing feature extraction network,this paper proposes a head pose estimation algorithm based on fine-grained and soft-stage regression.The algorithm first uses the anti-aliasing feature extraction network to extract the feature information of each stage,then uses the fine-grained structure mapping to obtain the multi-scale feature information,and uses the coarse-to-fine strategy soft stage regression method to further improve the model's estimation performance.This paper conducts comparative experiments on multiple public datasets to verify the effectiveness of the algorithm in improving the prediction accuracy and generalization ability of the model.
Keywords/Search Tags:head pose estimation, deep learning, anti-aliasing, multi-task learning, fine-grained structure
PDF Full Text Request
Related items