Controlling the welding process by monitoring the weld pool surface becomes more and more popular in robotic arc welding. In this paper, we propose a monitoring system to infer the P-gas metal arc welding (GMAW) weld pool geometry from the reflected laser lines, which are projected onto the specular weld pool and reflected onto a diffusive plane. The parallel straight lines are distorted according to the shape of the weld pool and thus contain the weld pool's shape information. Accurately computing the equations of the imaged laser lines in the world coordinate system is critical for the subsequent weld pool shape estimation. This paper focuses on accurately segmenting and identifying the reflected laser lines and novel image processing algorithms are proposed to fulfill this specific task. Experimental results verified the effectiveness of the proposed algorithms.