Effective car video retrieval using feature matching in a mobile video cloud

This paper presents a novel application that intends to search a specific scene/location from car videos recorded by driving recorders. The car videos are assumed to be collected via a mobile video cloud, supported by wireless broadband networks, e.g., a WiMAX or LTE 4G networks. The search algorithm adopts SIFT feature matching and text motion compensated matching to retrieve the most probable video segments. Our contribution is to provide both SIFT and text matching a separate confidence level, and efficiently combine them to identify the matched video segments according to the continuity of significant feature matching. The experimental results show that the proposed algorithm achieves high accuracy retrieval rate up to 96.65%.