In the last decade, with the increasing requirement for the quality of equipment, efforts have been devoted to the development of new real-time detection technics for weld joints quality in many industrial fields such as the nuclear, chemical and aeronautical. Welding is a complicated process, which is often affected by the welding process and environmen- tal uncertainties, and it is easy to produce welding defects such as overlap, pore, spatter and incomplete fusion. In fact, the welding quality control is carried out by a traditional destructive and non-destructive methods such as macrographs, ultrasonic testing and x-ray detecting. However, these methods have some limitations : x-ray detection can easily result in side-effects on the human body and ultrasonic testing is susceptible to the location, orien- tation and shape of the defect. In addition, the in-situ observation of the weld pool dynamics using non-intrusive instruments will give information about the stability and consequently evaluate the real-time welding quality. In this study, the objective is to analyze by using se- veral sensors (arc voltage, current, cameras, etc.), the influence of welding parameters on the weld pool behavior of a 316L stainless steel made with the GTAW process and a metal input. The installation of two cameras to observe the top side (oscillation, width) and on the other hand, the observation of the side (height, wetting, etc.) is necessary. In the other hand, an algorithm based on computer vision and image processing is developed including four steps : image acquisition, image preprocessing, weld pool features extraction and clas- sification. The weld joints were classified into four classes according to their intensity. In the second part of the algorithm, different machine learning methods (KNN, Random Forest. . . ) are tested for detecting and classifying weld joints on one hand, and on the other hand they are compared by analyzing performance scores (accuracy, time of calculations. . . ). Actually, image processing and machine learning combination is trained with extracted features to predict the classification results with a high level perspective to realize the real-time intelli- gent identification of weld surface defects.