| 基于梯度特征与3D卷积的视频不同跨度行为识别 |
 点此下载全文 |
| 引用本文：戴立新，赵德中，孙琦，胡春成，魏赛，王亮.基于梯度特征与3D卷积的视频不同跨度行为识别[J].计算技术与自动化,2025,(4):122-128 |
| 摘要点击次数: 9 |
| 全文下载次数: 6 |
|
|
| 中文摘要:为了优化视频不同跨度行为识别性能，提出了基于梯度特征与3D卷积的视频不同跨度行为识别方法。捕捉视频中的动态信息变化，计算梯度特征值并映射到梯度空间。建立了基于3D卷积的多时长特征融合模型，融合不同时间节点下的梯度特征值。结合人体骨骼行为概率、标签隶属度对比以及0阶和1阶梯度深度特征值，实现视频中的多种行为识别。实验结果表明，所提方法在各种视频环境和行为种类下，识别准确率高达95.2%，误识别率和漏识别率分别低至3.1%和5.4%，推理时间仅为37 ms/帧，证明了该方法具有较高的实际应用价值。 |
| 中文关键词:多特征融合 监控视频 梯度特征 像素变化率序列 梯度空间 |
| |
| Video Different Span Behavior Recognition Based on Gradient Feature and 3D Convolution |
|
|
| Abstract:In order to optimize the performance of video behavior recognition across different spans, a video behavior recognition method based on gradient features and 3D convolution is proposed. The dynamic information changes in the video had been captured, and the gradient feature values had been calculated and mapped into the gradient space. A multi-duration feature fusion model based on 3D convolution had been established to fuse the gradient feature values at different time points. By combining the probabilities of human skeletal actions, the comparison of label membership degrees, and the 0th-order and 1st-order gradient depth feature values, multiple types of behavior recognition in the video had been achieved. The experimental results show that the proposed method has a recognition accuracy of up to 95.2% in various video environments and behavior types, with a low false recognition rate and missed recognition rate of 3.1% and 5.4%, respectively. The inference time is only 37 ms/frame, proving that the method has high practical application value. |
| keywords:multi feature fusion surveillance video gradient features pixel change rate sequence gradient space |
| 查看全文 查看/发表评论 下载pdf阅读器 |
