行人检测一直是计算机视觉领域的热点和难点问题.本文提出了一种结合玻尔兹曼机 RBM(Restricted Boltzmann Machine)和支持向量机 SVM(Suport Vector Machines)的深度学习网络进行行人特征提取和分类,多层玻尔兹曼机无监督的训练网络参数得到行人特征并级联 SVM构建特征分类器进行特征分类,在融合多种行人数据库的基础上扩充了行人数据样本,满足深度学习对于大数据量样本的要求.实验中对比了不同层数网络对于模型性能的影响以及与传统人工特征相比在复杂场景下的行人检测效果,验证了深度学习对于行人特征提取的有效性.%Pedestrian detection has always been a hot and difficult issue in the field of computer vision.In this paper,a deep learning network that combines Restricted Boltzmann Machine (RBM)and Suport Vector Ma-chines (SVM)was proposed for pedestrian feature extraction and classification.The unsupervised training net-work parameters of Multilayer Boltzmann Machine were used to get pedestrian characteristics,which were combined with SVM to construct the feature classifier for feature classification.Based on the integration of vari-ous pedestrian databases,the pedestrian training samples were expanded to meet the requirements of deep learning for large data samples.In the experiment,the effects of the networks with different layers on the per-formance of the model were compared.The pedestrian detection effect of this method was compared with tradi-tional artificial features in complex scenes.The effectiveness of deep learning for pedestrian feature extraction was verified.
展开▼
