基于深度学习的监控视频目标检索

摘要

The content⁃based object retrieval is one of the most important research topics in video surveillance.This paper presents a novel object retrieval approach based on deep auto encoder.This approach takes advantage of the mask information to assist object rep⁃resentation,and introduces manual noise into the learning approach,which enhances the robustness of feature representation in the deep neural network.The experimental results prove the effectiveness and superiority of this approach.%基于内容的目标检索一直是视频监控领域最重要的研究内容之一。面向视频监控应用场景，提出了基于深度自动编码机的目标检索方法。通过在训练过程加入掩膜图像辅助信息和人工噪声，提高了深度神经网络特征表示的鲁棒性。实验结果证明了该方法在监控视频目标检索任务中的有效性和优越性。