[Purposes]The changes in road alignment observed by drivers are closely related to driving safety. Therefore，this paper aims to find reliable description indexes of alignment perspective characteristics to study the correlation between the changes of road alignment in drivers' vision and driving safety, and to quantify the impact of alignment perspective characteristics changes on driving safety by establishing road accident frequency prediction models.[Methods] This study utilized shape description tools to process driving images and obtained the shape center distance kurtosis and skewness of alignment perspective profile as describing indexes of alignment perspective profile, achieving parameterized expression of alignment shape changes from driving perspective; A Possion regression model was used to establish a highway accident frequency prediction model based on the alignment perspective profile index, which intuitively revealed the correlation mechanism between the changes in alignment perspective profile caused by road turning and accidents.[Findings] Empirical analysis was conducted on the 160 kilometer section of the Guangdong Sanshui to Huaiji section of the Second Guangzhou Expressway (G55). Results show that the mean absolute deviation (1.293) and cumulative residual (432.968) of the accident prediction model based on the center distance indexes are lower than those based on the traditional two-dimensional and three-dimensional alignment geometric indexes’ mean absolute deviation (1.302) and cumulative residual (434.694), indicating that compared with the traditional two-dimensional and three-dimensional alignment geometric index, the alignment characteristic index extracted from the driving perspective has better prediction accuracy for the frequency of traffic accidents and is more conducive to explaining the relationship between the alignment and the accident. The skewness index, which represents the symmetry of the profile, and the kurtosis index, which represents the sharpness of the profile, show a significant positive correlation with the accident frequency (p<0.01).[Conclusions] The method established in this study for quantitatively describing highway alignment from a driving perspective and the accident prediction model can provide a reliable reference for highway alignment optimization and accident prevention.