A method that combines facial dividing feature line and point heatmap regression is proposed to address the problem of low accuracy of face landmark detection caused by different facial feature types and scales in the cases of large posture changes and occlusion. A deep learning model based on two-stage stacked hourglass network is designed to realize feature analysis and landmark location. Based on the proposed method, the detection algorithm is developed, and the proposed method is compared with other methods by experiments based on several common image datasets. The experimental results show that the proposed method can adapt to the applications of large posture changes and face partial occlusion. Compared with other methods, the proposed method has less detection error and higher accuracy in face landmark detection.