Intelligent Live Broadcast Shot Switching System Based on Semantics

In modern society, live broadcasting has become a popular real-time interactive medium, covering a wide range of fields such as news conferences, sports events, concerts and education. In the traditional multi-camera live broadcast, the director needs to switch between different shots in real time according to the site conditions. This process is highly dependent on the director's individual judgment and reaction speed, which is under great pressure when facing complex and changing application scenarios. Therefore, this paper proposes a semantic based intelligent shot switching system. It combines real-time face recognition technology with keyword matching to achieve more accurate and smarter shot switching. First of all, we collected face photos of different genders and age levels as datasets, and preprocessed them. Then, we use the YOLOv8 model for face recognition. After training our customized datasets, we loaded the model to extract the character features from different camera screens. By comparing the similarity between face features and keywords, we calculated a matching score for each picture, and then selected the picture with the highest score as the main screen. In addition, this paper also designed a visual interface for the intelligent switching system based on PyQt5, which will provide users with a convenient use experience.