Drowsy driving is an emerging aspect of traffic accidents that causes serious injuries, fatalities, economic loss, and more around the world. This paper attempts to review all the drowsiness detection methods systematically and comparatively to facilitate the real-time, precise, and workable implementation of solutions. It covers traditional detection methods, including physiological indicators (e.g., eye blink rate, EEG signals, heart rate variability) and behavioral criteria (e.g., steering wheel movement, lane deviation), alongside newly developed techniques based on machine learning, deep learning, and computer vision. Our comparative analysis brings out that physiological methods are very accurate in closed settings but require sensors or wearables that are often invasive, thus limiting their practical use for everyday applications. Behavioral methods, on the opposite side, remain non-invasive and cheap; however, detection can be falsely triggered due to variability from factors such as road conditions or driving habits. In contrast, ML- and CV-based methods, as exemplified through facial landmark tracking, eye closure detection, and head pose estimation, provide perhaps the best compromise between accuracy and practicality, especially when combined with in-vehicle cameras and real-time data processing. However, the challenges related to data variability across individuals, environmental lighting conditions, and computational complexities persist for these sophisticated systems. The review delineates that hybrid systems, those merging more than one modality, e.g., from behavioral cues to computer vision, exhibit ever greater robustness and accuracy in real-world scenarios.