A magnetic resonance based dynamic imaging method comprises the steps that dynamic data is under-sampled by adopting TGRAPPA technology; the data obtained through under-sampling, and second under-sampling is performed by adopting a variable-density sampling acceleration mode; a second-time under-sampling signal is reconstructed; simulation parallel under-sampling is conducted on K space data obtained after reconstruction in the TGRAPPA mode, a full-sampling automatic calibration data ACS is established according to the under-sampled data; the full-sampling automatic calibration data ACS is mapped to a space higher than a current space in dimension, and the weight coefficient of each coil is calculated in a novel space; the K space data undergoing the simulation parallel under-sampling is filled according to the weight coefficient, and second time of reconstruction is perform to generate a final image. The magnetic resonance based dynamic imaging method does not need additional acquisition of automatic calibration signals from a K space center, noise amplification caused by acceleration time increase in parallel imaging can be well inhibited under high acceleration times, and images high in signal to noise ratio can be obtained.