Magnetic confinement fusion tokamaks are complex devices where a large amount of power is required to make the fusion reactions happen. In such experimental conditions, plasma facing components (PFCs) are subjected to high heat fluxes that can damage them. Machine protection functions must then be developed to operate current and future devices like ITER inthe safest way. In current tokamaks like Tore Supra, IR thermographic diagnostics based on image analysis and feedback control are used to measure and monitor the heating of the PFCs during plasma operation. The system consists of detecting a high increase of the IR luminance signal beyond fixed qualitative levels for a set of predefined regions of interest (ROIs). The detection of overheating regions is then fully dependent on the settings of the ROIs and of the qualitative thresholds. This ROI-based approach must be improved to fit with ITER requirements and operation, where the IR scene complexity (many components monitored at the same time) will be a real challenge for the real-time PFC protection. In this paper, we propose a new vision-based approach for the automatic recognition of thermal events. This ROI-free approach, which relies on intelligent vision system concepts, is composed of two main tasks: hot spot detection and thermal event recognition. We present the results of our approach for the recognition of one critical thermal event and compare its performance with that of the previous system.