Thermal cycles occurring during arc welding affect the mechanical properties of the welded parts. Fast heating and cooling cause high thermal gradients which induce plastic flow and, then, residual stresses. Metallurgical changes can also take place according to the material composition. Thermal cycles are a consequence of arc welding heat input. In order predict the thermal field, it is necessary to estimate accurately the thermal loading (heat source). In this work, a methodology is proposed for a fast estimation of heat source parameters based from non-intrusive data (weld pool contour). A surrogate model is established to link the weld pool contour to the heat source parameters. Thus the computational time required for the parameter estimation was significantly reduced in the optimization loop. This methodology is applied to the gas tungsten arc welding process on a thin stainless steel plate with a fully penetrated weld pool. The weld pool was observed on the back side with a camera in order to avoid electrical arc disruption.