This paper proposes a pyramid-based coarse-to-fine background-oriented schlieren (BOS) tomography method to improve three-dimensional reconstruction accuracy in fields with large density gradients. First, a pyramidal BOS digital system with multiple resolutions is constructed to enable synchronized upsampling and downsampling of both flow fields and background images. Subsequently, at each pyramid level, image warping and forward linear/nonlinear tracking are used to progressively correct the projection data and projection matrices, enabling tomography reconstruction at different spatial scales and high-resolution refinement. The performance of the proposed method was demonstrated in both synthetic and real BOS cases.