Aiming at the technical difficulties of the visual online inspection of large workpieces, an image mosaic method based on binocular vision feature point matching is proposed. Feature points are detected and matched based on an improved scale-invariant feature transform algorithm. A random sampling consensus algorithm is used to estimate the parameters of the transformation model to eliminate mismatched points, and a weighted smooth fusion method is used to fuse the spliced traces to complete the image splicing and fusion. The flexible visual detection system platform is built and the detection experiment of the workpiece feature area is carried out. The experimental datas are compared with the actual value of the workpiece feature area to verify the correctness and effectiveness of the mosaic method. Experimental results show that the proposed method meets the requirements of fast image stitching using a binocular-vision system in actual industrial production.