Research On Video Retrieval Technology Based On 3D Convolutional Neural Network

Content based video retrieval has been a research focus on video processing in recent years and content based video retrieval technology is studied in this paper.Firstly,study the structure of artificial video feature and extract the improved dense trajectory feature;then study 3D convolutional neural network based on Caffe framework and automatically learn video spatiotemporal feature through the network.The 3D convolutional neural network used in this paper consists of fifteen layers,including eight convolution layers,five maximum pooling layers,two fully connected layers.Measure respectively video similarity with improved Hausdorff distance for the features extracted by the two methods and get the video retrieval results.The experimental results show that 3D convolutional neural network and the improved Hausdorff distance proposed in this paper showed well performance in video retrieval.