Research On The Method Of Protagonist's Clothing Detection And Retrieval In Video Based On Deep Learning

With the popularity and development of the Internet,information technology has been fully integrated into the social life,"Internet plus" gradually enjoys popular support.As a concern for emerging applications in the Internet era,the quantity and traffic of the video industry are increasing rapidly,and the user groups are also expanding rapidly.While watching the video,items around the protagonist in the video gradually become a new fashion trend,which has attracted widespread attention,especially clothing on the protagonist.How to quickly and accurately detect the fashion clothing on main characters,and recommend similar clothing then link to the electricity business platform,which has become a popular focus of the audience.In order to reduce the cost of manpower detection and recommendation,and improve the viewing experience of the audience,this paper puts forward the research on the detection and retrieval method of the clothing on protagonists based on deep learning.Through deep learning,it realizes detecting and retrieving the clothing of the key roles in the videos.This method obtains recommended clothes and their purchase links,which can solve the problem of difficult to find clothing and improve the satisfaction of viewing experience.Visual chain technology is an important application of computational advertising,this paper establishes and implements an end-to-end clothing advertising chain system,which makes the video protagonist and clothes associated with the electronic business platform.This paper puts forward the protagonist verification algorithm based on video content,and summarizes the current development situation of person detection algorithm,posture selection algorithm,face detection and identification algorithm.It gradually reaches to clothing on protagonist of the video through a series of selection operation.This paper summarizes the development status of object detection field at home and abroad,and uses the convolutional neural network to extract the deep features.Combined with the advanced detection framework,this paper compares the effect of different network structure under clothes detection tasks to improve algorithm based on clothing feature.Finally,this paper realizes a clothes detection method under the video scene.This paper summarizes the current research situation of image retrieval based on text and content,and uses the triple loss function to extract feature to train retrieval network.It pays more attention to the feature difference of the heterogeneous data and the feature similarity of the same type of data,and proposes a trace ratio algorithm of the adaptive parameters as the condition of triplets' selection.In each iteration,as hard class samples,triplet with the largest distance ratio is chosen as the training of the model.Through this adaptive format with gradually strict standard,this paper realizes clothing retrieval model based on triplet loss.This paper uses deep learning to realize the detection of clothes worn on the main characters in the video,and uses a new method of clothes retrieval to achieve the similar clothes recommendation,finally forms a complete visual chain from video content to the business platform.The results show that the proposed method has rigorous selection for the identification of the protagonist in the video,and higher detection accuracy in the clothing detection problem,and also the advantages that the traditional methods do not have in the issue of clothing retrieval.