基于残差神经网络的视频内容快速检索系统研究

为解决公共安全视频图像信息轨迹追踪的问题，设计一个基于图像识别的视频内容快速检索系统，使用方向梯度直方图特征提取( HOG: Oriented Histogram Feature Extraction) 算法进行人脸定位，通过迁移学习对ＲesNet 模型进行再训练建立专用的神经网络的分类器。进行视频内容检索的过程是提取视频关键帧，对其中人物进行定位，提取其特征值，使用训练好的分类模型进行分类。将分类信息标注在图片上，将有关信息存入数据库，使用时查询数据库，得到详细信息。实验结果表明，该系统可有效地对视频内信息进行快速定位，对截取图片可以进行精准识别，在公共安全领域具有应用前景。

Video Content Quick Search System Based on Residual Neural Network

For public safety video image information trajectory tracking problem，a fast video content retrieval system based on image recognition is designed． The HOG( Oriented Histogram Feature Extraction) algorithm is used for face location，and the ＲesNet model is re-trained through migration learning to establish a dedicated Neural network classifier． The process of retrieving video content is to extract the key frames of the video，locate the characters therein，extract the feature values，and use the trained classification model for classifying． The classification information is marked on the picture，the relevant information is stored in the database，and the database is queried for detailed information． Experimental results show that the system can effectively locate the information of the video and accurately capture the image． In the field of public security，it has application prospects．