Real Time Shot Boundary Detection in Live 360 Degree VR Streaming Using Deep Learning

We suggest a real-time approach for detecting shot boundaries in live 360-degree VR streaming, utilizing deep learning techniques. Seamless transitions between shots are vital to providing users with an immersive experience in VR streaming. However, the current methods of shot boundary detection face issues in handling the real-time nature of live streaming and the distortions caused by 360-degree projection. To tackle these issues, we recommend a convolutional neural network (CNN) based method for identifying shot boundaries in 360-degree VR videos in real-time. This approach preprocesses the 360-degree VR video stream and feeds it to the CNN for detecting shot boundaries in real-time. We evaluated the suggested approach on a dataset of live 360-degree VR streams and compared its performance with existing shot boundary detection methods, including traditional techniques and deep learning-based approaches. The results of our experiment demonstrate that the proposed method delivers high accuracy and performs well in real-time shot boundary detection, surpassing the current methods. To demonstrate its robustness and versatility, we assessed the suggested method under different scenarios, such as different types of VR streams and network configurations. The suggested method can significantly enhance the user experience in live 360-degree VR streaming and can be beneficial for several applications, such as live events, sports, and concerts. Moreover, the proposed method can serve as a base for future research on shot boundary detection in VR streaming.