Autoclaves are commonly used pieces of equipment in fields, such as wet metallurgy and chemical engineering, and face the threat of leakage. Once a leak occurs, it can lead to unstable pressure inside the autoclave, and in severe cases, it can even cause explosions, posing a threat to production safety. In this regard, we propose a wet metallurgy autoclave leakage detection algorithm that combines infrared image filtering, segmentation, and frame difference methods to extract morphological changes in the leaking gas-phase medium. The real-time leak location was determined using a multivariate Gaussian analysis. Experimental results demonstrate that the proposed algorithm achieves a detection accuracy of 95% within a 1-second time span after a leak occurs and a detection accuracy of 98% within a 2-second time span, effectively pinpointing the precise location of the leak source.