Design And Implementation Of Fragment Recommendation System Based On Recognition Tracking

Increasing data makes it increasingly difficult for people to obtain effective information.In order to explore the value of data and take into account individual differences,recommendation systems have emerged.Using a trained algorithm model to select items that are most interesting to users from a vast amount of data based on their past behavior and records greatly saves time.At the same time,the popularity of smart TV is becoming increasingly widespread.not only the quality of TV programs is getting higher,but also the recommendation of TV dramas,movies,variety shows,and other programs is more in line with user preferences,greatly improving the popularity of smart TV.Under the current recommendation system,users can click on a movie system to recommend other movie and television works of the same theme type or create a collection of highlights,and users can upload their favorite movie and television clips on the webpage or APP client.However,most movie and television product recommendation systems in the industry do not deeply analyze user preferences,only staying at the level of recommendation for the entire movie,lacking analysis of actors and clips.In order to solve the shortcomings of the above recommendation system,this paper designs and implements a movie segment recommendation system,which combines current face recognition and target tracking technologies to recommend the highlights of a movie from the perspective of actors and a collection of other movie segments of the same genre that actors participate in.The specific work of the fragment recommendation system is as follows:(1)Face recognition part.This section is responsible for identifying the actors in the movie,and the movie enters the network after segmentation,frame extraction,and cropping.In the face recognition process,using Wiener filtering to improve clarity for image blurring issues,and using global equalization to improve image brightness and contour boundaries between actors and the environment for dim ambient light issues.(2)Target tracking section.This section uses the single target tracking network Siam RPN++to track actors in movie clips,and optimizes the tracking of small targets.Firstly,using data enhancement technology to generate more data on a limited dataset solves the problem of insufficient data samples;Secondly,from the perspective of target segmentation,add a mask branch in the network to improve the tracking and positioning ability of the network for small targets;Finally,the anchor frame selection mechanism is optimized to improve the problem of anchor frame redundancy.(3)Platform design part.The platform is divided into cloud and TV ends.The cloud end includes a cloud recommendation module and a time interval acquisition module.The TV end includes a fine-tuning module,interface display,and user behavior recording.The cloud recommendation module implements the preliminary screening of movie lists through recall and rough sorting algorithms,and the time interval acquisition module completes the acquisition of the time period during which actors appear in the movie;The TV side fine arrangement module selects the 10 movies that best match the user’s interests from the candidate list generated by the coarse arrangement algorithm;Finally,the system was tested and analyzed to verify that the segment recommendation system can achieve real-time and accurate recommendation of movies and segment collections.In this paper,the segment recommendation system based on face recognition,target tracking and recommendation algorithm ensures real-time and accuracy.The added segment set recommendation function provides technical support for the recommendation optimization of intelligent TV terminals,and has high application value and market value.