The invention provides a global detection method of unstructured outdoor terrains in the technical field of robots. The method comprises the specific steps of image acquisition, image preprocessing, scene image segmentation, appearance characteristic extraction, terrain type judgment, terrain classifier candidate database establishment, current scene optimal classifier verification, terrain type statistical modeling, statistical model parameter training and statistical model reasoning. According to the global detection method, the navigation experience of different time points is integrated in principle, and the stability and the intelligence degree of terrain detection under the condition of uneven and even deficient view field samples are improved; a combined multi-layer sensor classifier adopted for the global detection method takes the uncertainty and the nonlinearity of the appearance-terrain mapping relation of the scenes into consideration, so that the accuracy of terrain detection can be improved; in addition, the space relations at variable scales in the current scene and the compatibility of types are modeled in principle, and the global detection of the unstructured outdoor terrains is realized.