In response to the problem of low accuracy of human digital models leading to significant simulation calculation errors in the safety of human electromagnetic dose inside automobiles， the strategy to improve the accuracy of human digital models is proposed from three aspects. Firstly， a high-resolution medical imaging human digital model that is superior to traditional simplified self-built models is adopted. Then， fully considering the dispersion characteristics of the dielectric parameters of biological tissues with frequency variation， the dielectric parameters of biological tissues measured by different scholars abroad （taking liver and anisotropic muscles as examples） were compared using MATLAB， and domestic and foreign literature data were compared to verify the accuracy of the measurement data. Finally， the smoothed rotation enhanced of as-rigid-as-possible （SR-ARAP） algorithm is introduced to transform the upright posture of the human digital model into a sitting posture， which matches the actual parameters of a certain type of car seat. The results indicate that the digital model of the human in medical imaging has high resolution of various organs and tissues， with around 2.1 million units in the 50th percentile model. The measured dielectric parameters of biological tissues by different scholars show good agreement in the mid-frequency range but poor consistency in the low-frequency and high-frequency ranges. Compared with the traditional as-rigid-as-possible （ARAP） algorithm， SR-ARAP improves the collapse of popliteal fossa formation， and the volume change of the main tissue in SR-ARAP is mostly less than 50% of ARAP， which has the advantage of low deformation distortion.