Research On Head Pose Estimation Method Based On Computer Vision

With the widespread use of artificial intelligence and human-computer interaction technology,the head pose detection method based on computer vision has also become the key research direction of computer vision technology because of its convenient and comfortable natural interactive experience.It has been widely used in the fields of intelligent monitoring,fatigue driving detection,virtual reality and so on.The traditional head pose detection algorithm is easily affected by the interference of background noise,illumination change and local occlusion in the natural environment.The accuracy and real-time performance of the algorithm cannot meet the requirement in practical application.Under this background,the head pose detection method in the natural environment is studied and improved in this thesis.The video of head motion is taken by monocular camera,and the algorithms are used to analyze and process the captured two-dimensional plane image.The head pose angles of the human head in the 3D world coordinate system is obtained via the analysis and process of 2D image.The proposed method of head pose detection mainly focuses on the following four aspects.1.A face detection method based on cascaded convolution neural network is proposed.The network is composed of two deep convolution neural networks,and the two level network is used to filter the detected face bounding box from coarse to fine.During the training process,the inner connection of the face detection and landmark detection are used to integrate the face detection data with the face feature points in the same network's training data.This method can improve the accuracy of the face detection.The experiment shows that the method has good robustness to light change,partial occlusion,head posture change and expression change in natural environment,and has a high accuracy of face detection.2.To solve the problems of large amount of computation and slow detection speed in the traditional face detection process,a detection based inter frame tracking method is proposed.The Gauss kernel correlation filter is introduced to simplify operation and improve detection speed.The feature of the local two value pattern is extracted from the image to improve the representation ability of the face feature,and the long time tracking of the target is guaranteed by setting the appropriate tracking threshold and updating the model online.3.Based on the practical application background of this research,through the analysis and comparison of the mainstream face feature point marking method,the constrained local model algorithm with high accuracy is chosen as the algorithm of the face alignment step in this thesis.Then the Pn P algorithm is used to calculate the posture angles of the head,and it is transmitted to the hardware platform to control the rotation of the remote cloud platform.The remote head servo system based on the head pose detection is realized.4.The authoritative dataset is used to test the algorithm designed in this thesis.The test results show that the head pose detection method in this master's thesis has high accuracy and speed.by designing reasonable experiments in the natural environment,it is proved that this method can detect head posture steadily,and it has certain feasibility and applicability.The existing problems of the system are analyzed,and the follow-up work is prospected in the thesis.