Research On Video Character Recognition Algorithm Based On Deep Learning

With the rapid development of mobile Internet and mobile communication technology,mobile Internet has penetrated into all aspects of people's daily life,and more and more video data is accumulated on the network platform every day.How to retrieve relevant people from the massive video data is very important.It is often a very difficult task to simply observe and filter with the naked eye without using any algorithm.Based on the PyTorch deep learning framework and the iQIYI-VID-2019 data set as experimental data,this paper proposes a multi-modal feature fusion-based video person identification framework,fully analyzes and verifies its feature fusion details.The main work is as follows:(1)Discussed and constructed an algorithm framework for video person identification problem based on static image methods,processed millions of image data and extracted facial feature vectors,and designed experiments to verify the effectiveness of this feature;(2)Multi-modal feature fusion was introduced into the framework of video person identification algorithms,a video person identification model based on multi-modal features was constructed,and a variety of fusion algorithms were tested to alleviate the problems caused by lacking of data and noise problems.The experimental conclusions prove that the multi-modal video person recognition framework proposed in this paper can well integrate the multi-modal semantic features under different noise levels,overcome the accuracy reduction problem caused by the lack of some modal features.The results of any single-modal prediction have been greatly improved,fully verifying the effectiveness of the framework proposed in this paper.The algorithm proposed in this paper achieves 91.14%mAP in the iQIYI-VID-2019 test set.