In the paper, a near-infrared-ray (NIR) and side-view video based low-complexity drowsy driver detection system is developed for day and night applications. The proposed system detects drowsy conditions effectively whether or not glasses. To reduce the redundant computations, the pre-defined ROI (region of interest) is used at the procedures of face, glasses bridge, eyes, and nose feature detections. By the geometric based facial image processing, the eyes and nose positions are recognized, and the closed eyes and nod situations due to drowsiness are detected effectively. After simulating by self-made driver's video datasets, the average open/closed eyes detection accuracy rates without/with glasses are 95% and 84% in day, and are 89% and 87% in night, and the average drowsy detection accuracy is up to 90%. By software optimizations, the processing speed operates up to 150fps and 20fps in PC and the embedded system respectively for real-time drowsy driver detections.