Accidental fall is the most prominent factor that causes the accidental death of elder people due to their slow body reaction. Automatic fall detection is then an emerging technology which can assist traditional human monitoring and avoid the drawbacks suffering in health care systems especially in dusky environments. In this paper, a novel fall detection system based on coarse-to-fine strategy is proposed focusing mainly on dusky environments. Since the silhouette images of human bodies extracted from conventional CCD cameras in dusky environments are usually imperfect due to the abrupt change of illumination, our work adopts thermal imager instead to detect human bodies. In our approach, the downward optical flow features are firstly extracted from the thermal images to identify fall-like actions in the coarse stage. The horizontal projected motion history images (MHI) features of fall-like actions are then designed to verify the fall by the proposed nearest neighbor feature line embedding (NNFLE) in the fine stage. Experimental results demonstrate that the proposed method can distinguish the fall incidents with high accuracy even in dusky environments and overlapping situations.