Research On Clothing Recognition Method Of Specific Characters In Video

In recent years,with the booming Internet economy,online video traffic has grown dramatically,and the video user community is rapidly expanding.At the same time,online video advertising revenue has also increased dramatically,making video advertising a huge potential business opportunity in the online video market,attracting more and more research to develop new advertising models for the media.The same costumes in video have gradually become the focus of widespread concern and chasing.In order to make advertisements closely related to video content,reduce the interference to video users and improve the efficiency of advertising promotion,we need to study more effective of advertising,and take advantage of the traffic effect brought by stars.In order to combine the video website and e-commerce two major Internet platforms and convert user traffic into product sales,this paper proposes a clothing fashion data mining method based on star recognition.This method aims to identify the clothes of a specific person in the video,such as the protagonist,and get the category of clothes and the pieces of clothes that the person wears in the video.This topic mainly includes two parts:(1)Clothing detection of specific characters in the video,including human body area detection,pose selection,face detection and identity verification,and clothing detection;(2)The identified clothing images are clustered,and the similar clothing images detected by the continuity of the video frames are classified into one category,and the cluster center is used as a representative of the clothing to remove redundant similar data.This paper first summarizes the development status of target detection,image feature extraction,image similarity judgment and clustering algorithm at home and abroad,and compares and analyzes different target detection models.Then,for the detected clothing image,the image feature is extracted using a depth model based on the triple loss function.The method directly obtains the mapping from image to a compact European space through end-to-end learning and the distance in the feature space is directly used to represent the similarity between clothing.Finally,in the learned feature space,the classical density clustering algorithm is improved,and a clustering algorithm suitable for this scene without specifying the number of clusters is proposed.Experiments on real video datasets show the feasibility and superiority of the proposed method.