The reconstruction of flow field density from the measured value of light deflection angle passing through the disturbed flow field, it is great significance for the in-depth understanding of complex flow phenomena. Based on the deflection angle of light passing through the disturbed flow field obtained by the wind tunnel video measurement method, the reconstruction method of complex flow density is studied. According to the relationship among density, refractive index, and light deflection angle, the partial differential equation of density with prior information is derived by variational method. The artificial neural network is used to model the measured data, and the error magnification caused by the loss of local peak value is effectively suppressed. The results of numerical simulation and wind tunnel experiments show that the density reconstruction result of this method is obviously better than that of the direct partial derivative method, especially at the peak, this method has higher accuracy in terms of value, which provides a new way for the quantitative analysis of complex flow structure.