一种基于前馈神经网络的视频动作质量评估方法及装置
By introducing a pre-trained video feature extraction network and a multi-layer feedforward neural network based on residual structure, the problem of insufficient feature learning in video motion quality assessment is solved, achieving higher accuracy and stability in motion quality assessment, and possessing the advantage of lightweight design.
CN118521939BActive Publication Date: 2026-07-17SHANGHAI UNIV
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI UNIV
- Filing Date
- 2024-05-29
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
Traditional video motion quality assessment methods suffer from insufficient video feature learning capabilities, resulting in poor assessment accuracy and stability.
Method used
A pre-trained video feature extraction network is used to extract preliminary features, and a multi-layer feedforward neural network based on residual structure is used to aggregate and learn the preliminary features. The action quality score is then predicted by combining the score distribution regression module.
Benefits of technology
It improves the accuracy and stability of video motion quality assessment, realizes a lightweight automatic scoring method, and has faster convergence speed and less training time.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN118521939B_ABST
Abstract
本发明涉及一种基于前馈神经网络的视频动作质量评估方法及装置。该方法首先获取原始动作质量评估视频并进行降采样，得到多个视频片段，利用数据增强策略扩展视频片段，得到相应的图像帧；其次，预处理图像帧；接着，利用预训练的视频特征提取网络处理预处理后的图像帧，获取动作质量评估视频的初步特征，初步特征包括视频中动作的时空语义信息；再次，利用基于残差结构的前馈神经网络对初步特征进行聚合与学习，得到增强后的特征；最后利用分数分布回归方法对增强后的特征进行预测，得到动作质量分数预测结果。与现有技术相比，本发明具有有效提高视频动作质量评估的准确性和稳定性等优点。
Need to check novelty before this filing date? Find Prior Art