The invention relates to an inertia/visual integrated navigation method adopting iterated extended Kalman filter and a neural network, belonging to the technical field of integrated navigation in a complicated environment. The method comprises the steps of when a visible signal is valid, acquiring a dynamic video by utilizing a camera carried by a mobile robot, and determining the speed of the camera by an image characteristic extraction method and a nearest neighbor matching method; optimally estimating the speed and the acceleration of the mobile robot by using the iterated extended Kalman filter; establishing a navigation speed error model of an inertial navigation system by utilizing the neural network; when the visible signal is in loss of lock, compensating the speed error of the navigation system by virtue of the neural network error model which is previously obtained by training. According to the method, the problem that the inertia/visual integrated navigation system can not provide lasting high-precision navigation when the visible signal is in loss of lock can be solved; the method can be applied to long-endurance, long-distance and high-accuracy navigation and location for the mobile robot in the complicated environment with weak light, no light or the like.