Study On Hand Pose Based On Complex Scenes

Gesture is a very important means in human-computer interaction.Because of the rapid development of the artificial intelligence technology,augmented reality and virtual reality technologies,and the popularity of depth cameras in recent years.Real-time and accurate hand pose estimation has become a hot research topic in the field of artificial intelligence,it plays an important role in our daily life,computer vision provides a new and better idea for the research and development of hand pose estimation,and 3D hand pose estimation is also a challenging task in computer vision.In recent years,the convolutional neural networks have been widely used in computer vision and have achieved great success.At the same time,with the rapid development of the Internet,various new large public standard datasets keep appearing,which promotes the development of the convolutional neural network in the field of hand pose estimation greatly.To solve the problem of complex target scene and high degrees of freedom of hand in the field of hand pose estimation,a 3D hand pose regression model based on the convolutional neural network structure of attention mechanism is designed in this paper.The main research work of this paper are as follows:In order to solve the problem that the background in complex scenes has an impact on hand pose estimation,the depth image which contains the human hand is segmented precisely from the complex scenes of original depth image according to the depth value information.Aiming at avoid the influence of inconsistent distance between the target and the camera plane,which affects the accuracy of hand pose estimation,we normalize the depth image.At the same time,in order to locate the region of interest,we design a convolutional neural network based on attention mechanism to solve the hand pose estimation problem.In order to train the convolutional neural network model faster,we reduce the dimension of the groundtruth by applying the principal component analysis method,which reduces the parameters of the full connection layer greatly in the convolutional neural network,it was also equivalent to adding the constraint of hand pose for the hand,and comparative experiments are conducted to verify the effectiveness of the loss of heatmap.The experimental results in this paper show that the loss of heatmap can improve the generalization performance of our proposed neural network model,and that the convolutional neural network model based on the attention mechanism achieves better performance on three public standard datasets,which proves that our method is effective and superior.