Multi-view Video Coding with Decoupled Neural Representation for Multi-modal Traffic Data

With the continuous development of perception and analysis technology, intelligent transportation has become an emerging research field. The analysis and decision-making of traffic scenes often require massive multi-view and multi-modal data, which brings great challenges to storage and transmission. Recently, video compression methods based on implicit neural representation have shown great potential. However, the existing works are not enough to resolve the correlation and information redundancy of multi-view videos, resulting in poor compression performance. We propose a multi-view video compression method based on decoupled implicit neural representation for multimodal data(image intensity and depth) of traffic scenes. By decoupling multi-view video into camera index, video frame index, and spatial coordinates, we extend single-view video representation method to realize multi-view video representation and compression. The proposed method can effectively exploit the redundant information between views to better represent and compress multi-view videos. We conduct experiments on MPEG-I visual test materials, and the results show that our reconstruction quality is better than existing methods.