Obstructive sleep apnoea-hypopnoea syndrome (OSAHS) is a common sleep disorder characterised by repeated episodes of upper airway obstruction during sleep. The gold standard for diagnosis is in-laboratory polysomnography (PSG), which is limited in both availability and accessibility. This limitation motivates us to develop an alternative diagnostic tool that automatically analyses short (between 8 and 12 min) video recordings captured using a mobile phone. Based on clinical studies, we focus on facial features, specifically the openness of the eyes and mouth. Off-the-shelf facial landmark detection models are used to estimate the perimeters of the eyes and mouth, and using a proposed relative openness (RO) metric, we measure mouth and eyes openness in each frame. The openness temporal signals are filtered and used to extract high-level features that can assist in automated OSAHS diagnosis. We gathered an annotated dataset of twelve short videos, with a total duration of 122 min, recorded during children’s sleep in real world conditions including low quality recordings, low light, occlusions, abnormal body postures and partial views. Using this dataset, we conducted two evaluations: estimation accuracy, by comparing our results against clinician-annotated eye and mouth states, and predictive validity, by comparing the extracted features to the gold standard Apnoea-Hypopnoea Index (AHI). The proposed method demonstrates strong estimation accuracy, achieving 92.2% accuracy in detecting mouth open states, with precision and recall both around 84%. It also exhibits promising predictive validity, as features such as \(mouth\textrm{STD}\) and \(eye\textrm{SE}\) , as well as a linear combination of \(mouth\textrm{CAVG}\) , \(mouth\textrm{OAVG}\) , and \(mouth\textrm{SR}\) , show notable correlation with AHI. These results highlight the potential of video-based methods for assisted diagnosis of OSAHS.