Deflagration-like combustions pose a serious threat in various (e.g. industrial) scenarios. Detecting, locating and distinguishing these in their developing phase may significantly reduce the resulting damage to material and individuals. In this work, an image processing based algorithm for detection, localization and volume estimation of deflagrations with a multi-camera-system is proposed. The proposed algorithm has been tested with image sequences of real deflagrations as well as possible false alarm scenarios. In comparison to state of the art methods, false alarm safety, localization quality and robustness to noise have been improved significantly.