一种手部追踪方法、装置、追踪设备及存储介质
This hand tracking method, trained with a lightweight convolutional neural network and a specific loss function, solves the problems of deformity recognition and edge computing applicability in hand tracking models, achieving efficient and accurate hand tracking.
CN115880773BActive Publication Date: 2026-07-17SHENZHEN UNIV
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENZHEN UNIV
- Filing Date
- 2022-11-29
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
Existing vision-based hand tracking models suffer from problems such as deformity recognition due to restrictions on the degrees of freedom of hand movement and inapplicability of complex structures to edge computing scenarios.
Method used
A lightweight convolutional neural network is used for hand image feature extraction and mapping. A second lightweight convolutional neural network is trained by combining a first loss function based on spatial loss and a second loss function based on Gaussian process to improve the accuracy and rationality of hand key points in three-dimensional space.
Benefits of technology
It improves the accuracy and efficiency of hand tracking, is suitable for edge computing scenarios, and reduces computing costs.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN115880773B_ABST
Abstract
本发明适用计算机技术领域，提供了一种手部追踪方法、装置、追踪设备及存储介质，该方法包括：使用第一轻量级卷积神经网络对采集到的手部图像进行特征提取，得到包含手部位置信息的二维特征图像，使用第二轻量级卷积神经网络对二维特征图像进行映射处理，得到二维特征图像中手部关键点对应的三维坐标，该第二轻量级卷积神经网络由基于空间损失的第一损失函数和基于高斯过程的第二损失函数训练得到，从而提高了第一轻量级卷积神经网络和第二轻量级卷积神经网络的手部追踪准确度。
Need to check novelty before this filing date? Find Prior Art