This paper presents a software-hardware system for seam tracking during scanning laser welding, which is necessary to increase productivity and decrease defects, caused by manual homing. A vision sensor system has been designed for laser scanning head, which can acquire clear image of work region in a protective atmosphere. By analyzing the features of captured images a set of algorithms has been proposed to detect the edges of welding joint and extract the characteristic parameters for welding trajectory generation. System testing was conducted during the welding of titanium heat exchanger. The accuracy and reliability of developed method was measured and described.