Rapid manufacturing technologies have made it possible to reduce material wastes and to remanufacture valuable dies and tools. This paper focuses on reasonable utilization of materials and energies in gas metal arc welding (GMAW) for rapid manufacturing. During the weld-based additive manufacturing process, geometries of the deposited weld beads should be monitored and controlled. Using a composite filtering technique, a computer vision-sensing system was designed. Features of the weld bead image were analyzed. A corresponding image processing technology was used to extract parameters of the deposited weld beads. An on-line control of the deposited beads was realized based on a segmented neuron self-learning controller. The results show that the proposed control system is capable of keeping the deposited bead width of a thin-walled part consistent, making an efficient use of materials and energies possible.