Vision-centric BEV perception has recently received increased attention fromboth industry and academia due to its inherent merits, including presenting anatural representation of the world and being fusion-friendly. With the rapiddevelopment of deep learning, numerous methods have been proposed to addressthe vision-centric BEV perception. However, there is no recent survey for thisnovel and growing research field. To stimulate its future research, this paperpresents a comprehensive survey of recent progress of vision-centric BEVperception and its extensions. It collects and organizes the recent knowledge,and gives a systematic review and summary of commonly used algorithms. It alsoprovides in-depth analyses and comparative results on several BEV perceptiontasks, facilitating the comparisons of future works and inspiring futureresearch directions. Moreover, empirical implementation details are alsodiscussed and shown to benefit the development of related algorithms.