In　this　paper,　we　propose　a　projective　transformation　based　virtual　reference　frame　generation　scheme　for　video　coding　in　moving　camera　scenarios.　The　generated　virtual　reference　frame　can　provide　high　qualitative　reference　in　the　process　of　removing　temporal　redundancy　and　improve　the　compression　performance　of　the　encoder.　We　also　propose　a　depth　estimation　algorithm,　which　combines　optical　flow　with　the　triangulation　method　from　Simultaneous　Localization　and　Mapping　(SLAM),　to　acquire　the　accurate　pixel-level　image　depth　for　projective　transformation.　To　further　improve　the　compression　efficiency,　we　modify　the　syntax　of　High　Efficiency　Video　Coding　(HEVC)　to　support　the　coding　of　camera　parameters　and　new　motion　vector　scaling　method.　Experimental　results　show　that　compared　with　HEVC　reference　software　HM16.7,　our　method　has　a　Bjontegaard　Delta　bit　rate　(BD-rate)　gain　of　7.80%　and　a　encoding　time　gain　of　23.9%　in　spite　of　increasing　decoding　complexity.　©　2022　IEEE.