Depth image gesture estimation method based on semi-supervised learning

Abstract

The invention discloses a depth image gesture estimation method based on semi-supervised learning. Compared with an RGB image, the hand posture with higher precision can be estimated from the depth image. Although a current gesture estimation method based on deep learning achieves a good effect, the method depends too much on training by using annotation data and has a very complex process of annotating the three-dimensional gesture in the image. According to the invention, an efficient point cloud expression mode is provided, local features and global features are effectively fused, and a novel method for estimating the three-dimensional hand posture from the depth image with high precision is realized. By reducing the dependence on the annotation data in the model training process, the data annotation cost is reduced. Compared with the previous semi-supervised learning method, the method has the advantage that the breakthrough of hand posture estimation in precision is realized on the premise of ensuring the operation efficiency.

Description

technical field [0001] The invention belongs to the technical field of digital image recognition, and more specifically, relates to a method for estimating gestures of depth images based on semi-supervised learning. Background technique [0002] Automatic real-time 3D gesture estimation has received a lot of attention this year, which includes a wide range of application scenarios, such as human-computer interaction, computer graphics and virtual / augmented reality, etc. After years of intensive research, 3D gesture estimation has achieved remarkable progress in accuracy and efficiency. Since convolutional neural networks perform well in processing images, most gesture estimation methods are based on convolutional neural networks. Some methods use 2D convolutions to process depth images, but the features extracted by 2D convolutional neural networks are not suitable for direct 3D pose estimation due to the lack of 3D spatial information representation. To better capture the...

AI Technical Summary

Problems solved by technology

Aiming at the problem that a large amount of labeled gesture information is required in the training process of the gesture estimation model, but the cost of labeling 3D gesture labeling information is high, the present invention designs a semi-supervised deep network architecture, the purpose of which is to use unlabeled data and relatively Use less labeled data to train the entire network to improve the accuracy of 3D gesture estimation and reduce the cost of training data labeling

Images