The three-dimensional reconstruction technology of asphalt pavement textures based on image processing has the advantages of being fast, comprehensive, and high-resolution. Therefore, we propose a binocular vision technology based on multiscale image fusion aimed at the low accuracy problem of three-dimensional texture detections caused by the concentrated color of asphalt pavements and the lack of prominent feature points. First, we implemented multiscale decomposition of the image via weighted least squares filtering. Next, we fused the multiscale image information with a cross-scale aggregation model to three-dimensionally reconstruct the asphalt pavement, improving the accuracy of the binocular reconstruction. The proposed method was compared with a laser scanner through regional texture parameters. The results show that the regional texture parameters obtained by the proposed method are close to those obtained by the laser scanner.