This paper presents a new shared-control approach for brain-actuated intelligent wheelchair by means of a noninvasive Brain-computer Interface (BCI). The problem caused by the sparse and unsteady feature of BCI command, a two-layer shared-control strategy is proposed to steer the intelligent wheelchair. The first one is a machine decision layer which responsible for enabling/disabling the BCI command in a certain context, such as bifurcations and multiple-directions caused by new obstacles in the environment or deadlocks. The second one is a human intention matching layer which is used to generate suitable motion command with consideration of the human user's ability of driving the wheelchair, as well as the situation awareness of potential directions in a known environment. To achieve efficient navigation and position under condition of decoding uncertainly of BCI, the paper provides a navigation system to validate user's commands. And a steady state visual evoked potential (SSVEP) of the BCI as the human machine interface (HMI), the canonical correlation analysis (CCA) is applied to analyze the frequency components of SSVEP in electroencephalogram(EEG). Experiments have been performed by a number of able-bodied volunteers in a structured known environment. The experiment results show that all volunteers are able to successfully operate the wheelchair with a high level of robustness.