| 基于改进残差网络的驾驶员异常情绪识别方法 |
 点此下载全文 |
| 引用本文：潘东杰1，秦晓1，崔玉霞1，王宪伦1,2.基于改进残差网络的驾驶员异常情绪识别方法[J].计算技术与自动化,2025,(2):121-127 |
| 摘要点击次数: 886 |
| 全文下载次数: 751 |
|
|
| 中文摘要:为了保障交通安全，避免负面情绪影响驾驶员状态导致交通惨剧的发生，提出了一种结合注意力算法的改进残差网络来监测车辆驾驶场景中驾驶员的异常情绪。首先，使用自适应直方图等一系列手段对图像进行预处理，突出表情特征；然后，采用ECA块自优化通道的权重值，实现局部跨信道交互，提高精度，增强信道特性；最后，将注意力机制与特征提取算法融合实现表情识别。结果表明，与改进前的网络相比，该表情识别方法参数更少，速度更快，分类效果更好，在CK+、FER2013和RAF DB数据集上的准确率分别为96.38%、72.25%和87.97%。该算法能够更有效地识别驾驶员的异常情绪，对预防交通事故与保障行车安全具有积极意义。 |
| 中文关键词:深度学习 计算机视觉 表情识别 行为检测 |
| |
| Driver Abnormal Emotion Recognition Based on Improved RepVGG |
|
|
| Abstract:To ensure traffic safety and avoid negative emotions affecting the driver’s state leading to traffic accidents, this paper proposes an improved residual network combined with attention algorithms to monitor the abnormal emotions of drivers in vehicle driving scenarios. Firstly, a series of techniques such as adaptive histograms are used to preprocess the image and highlight facial features; then, using ECA blocks to self optimize the weight values of channels, achieving local cross channel interaction, improving accuracy, and enhancing channel characteristics; finally, the attention mechanism is integrated with the feature extraction algorithm to achieve expression recognition. The results show that compared with the pre improved network, this expression recognition method has fewer parameters, faster speed, and better classification performance. The accuracy on the CK+, FER2013, and RAF DB datasets is 96.38%, 72.25%, and 87.97%, respectively. This algorithm can more effectively identify abnormal emotions of drivers, which has a positive significance for preventing traffic accidents and ensuring driving safety. |
| keywords:deep learning computer version facial emotion recognition behavior monitoring |
| 查看全文 查看/发表评论 下载pdf阅读器 |
|
|
|
位访问者,今日访问
