Research on Video Fusion Algorithm of Rail Transit Station Based on Two-Stream Network

This paper proposes a video fusion algorithm based on a two-stream network for monitoring and securing rail transit stations based on the issue of large amounts of data, low resolution, and a single perspective. This method is intended to tackle the problems associated with large amounts of data, low resolution, and a single perspective. First, a video fusion framework is built using a two-stream network model to capture the video’s optical flow and feature information such as appearance, motion, and scene. Second, to eliminate the distortion and blurring that may occur in the process of video fusion, an adaptive weight allocation strategy is proposed. Third, according to the spatio-temporal characteristics of the station video, an optimization model is built that more accurately describes the video fusion effect considering the effects of temporal continuity and spatial coherence. Finally, a video fusion system of rail transit stations based on a two-stream network algorithm is built. Simulation and experimental verification prove the effectiveness of the proposed video fusion algorithm. To obtain an accurate video fusion model, it is necessary to study the characteristics of video two-stream networks, which lays a solid foundation for completing the extraction of video information and improving video quality as a result.