Near-infrared spectroscopy (NIRS) is a non-invasive neuroimaging technique that recently has been used to measure changes in cerebral blood oxygenation associated with brain activity. Numbers of research groups have applied general linear model (GLM) based method to analyze the NIRS data. However, classical GLM based method cannot provide on-line analysis. Therefore, its usage is constrained in processing NIRS data where real-time feedback is required. In the present paper, we are proposing a framework for NIRS based on-line brain activation mapping. The framework employs an extended GLM with coefficients updated by an extended Kalman particle filter for on-line brain activation mapping. A set of data recorded in a finger tapping experiment was studied using the proposed framework. The results so obtained, suggested that the method can effectively locate brain activation areas on-line with the noisy NIRS signal, thereby demonstrating its potential for real-time NIRS-based brain imaging applications.