基于改进SSD的人脸特征检测算法的研究

Research on face feature detection algorithm based on improved SSD

进行了基于面部特征识别的驾驶员疲劳检测研究。设计了一种基于改进SSD的人脸特征检测算法,引进Resnet网络残差模块,解决了传统SSD算法训练的网络退化问题;使用Resnet-50替换原始SSD网络框架,有效减少了模型参数并提高识别率。针对驾驶室背景复杂的应用场景,采用Focal Loss代替Softmax Loss,以平衡正负样本的比例。实验结果表明:改进SSD算法在自建的人脸特征数据集上的平均识别准确率为98.1%,比传统SSD算法提高了7.6%,体现为提高了对遮挡目标的识别率,且保持了较快的收敛特性。

In order to improve the safety of transportation,the study of driver fatigue detection based on facial feature recognition is carried out.This paper designs a face feature detection algorithm based on improved SSD.Resnet residual module is introduced to solve the network degradation problem of traditional SSD model training.Replacing the original SSD network framework with Resnet-50 effectively reduces model parameters and improves recognition rate.To reduce the interference of cab background,Focal Loss is adopted instead of Softmax Loss to balance the proportion of positive and negative samples.The experimental results show that the average recognition accuracy of the improved SSD model on the self-built face feature data set is 98.1%,which is 7.6%higher than that of the traditional SSD algorithm,reflectes the improved recognition rate of the occlusion target and the fast convergence characteristic.