In order to cooperate with the development of advanced driver assistance systems (ADAS), this research focuses on by using facial landmark features to evaluate the driver's concentration level and then determine whether the driver is in a safe driving state. Among the many existing face landmark detection technologies, this study conducts a comprehensive evaluation of multiple methods, including technologies, such as OpenCV CNN, PFLD, and Dlib. These methods have their advantages in certain application scenarios, but in driving assistance systems, especially in practical applications when the driver's head changes at different angles, their detection accuracy still have shortcomings, especially in facial deflection, or the recognition effect is not stable enough at different angles. To solve this problem, in this study, YOLO based technology is introduced for face landmarks detection and it reveals high detection accuracy when dealing with multi-angle changes in the driver's face. Compared with the traditional methods, YOLO based design maintains stable detection results at different angles, which provides more reliable technical support for the applications of ADAS. By combining the detection of facial feature points further, especially the eyes, YOLO achieves more accuracy to detect driver’s fatigue and concentration. Since the eye status is one of the core indicators for fatigue detection, the detection of eye feature points is optimized. The results show that after the combination with YOLO based technology, the accuracy of face detection and eye feature points positioning reaches 88.23%, significantly exceeding the performance of the other methods, including OpenCV CNN (i.e. 55.29%), PFLD (i.e. 60.52%) and Dlib (i.e. 50.18%). By experiments, the results show that the strategy of combining YOLO with facial landmarks detection greatly improves the accuracy of driver’s facial feature points detection for the ADAS applications.