DashCam video compression using historical data
Biao Ma, Amy R. Reibman · 2016
While dashcam videos (DCVs) are used to document unanticipated situations such as accidents, they are often used to create a training set for vehicle detection and vehicle behavior modeling. This requires a substantial volume of stored DCVs. In this paper, we propose a system that effectively compresses these DCVs by taking advantage of existing historical data. First, our video retrieval and alignment preprocessors construct a reference video for a new DCV based on GPS information and ORB (Oriented FAST and Rotated BRIEF) features. The 3D-HEVC encoder jointly compresses these two videos. Our illumination matching algorithm makes the system robust over different illumination conditions. Our system reduces the bit-rate around 30% when tested over 80 sequences and 3 different scenarios: highway, boulevard and downtown.