In this paper, a new method is proposed to estimate crowd density based on binocular vision: according to the average parallax of the object in the two camera plane, first we calculate the correction parameters of monitoring location, then we fit the correction function based on the result of the correction parameters of the same target objects in different locations. By taking foreground pixels, foreground edge pixels of image as features of crowd density estimation, we could perform parameter correction. Experiments show that by comparing with existing methods, this method is effective in reducing the projective deformity and can improve the effectiveness of the features greatly. So it could enhance the accuracy of the estimation of population density.