Automatic Identification and Prediction of Anatomical Points in Monocular Images for Postural Assessment

The postural assessment of an individual can be related to the angles generated from the markers of two bone references, at any time during routine movements, such as walking, sitting and standing. Postural evaluation assistance systems are commonly developed from the analysis of the gait. However, in this study an algorithm was developed based on activities of sit-to-stand as these activities are pre-requisite for the other daily activities. Based on this context the objective of this study is to develop an automated algorithm to identify a group of nine anatomical landmarks, using a postural assessment protocol of the sit-to-stand and stand-to-sit activities from a lateral view, allowing the extraction of information necessary for the protocol anytime during the execution of the activity. The proposed algorithm employs digital image processing techniques such as image segmentation and the prediction of the occluded points for identification of anatomical landmarks in patients through reflective markers. The results obtained show that the algorithm has an accuracy of 95.1% for the angular values calculated from the obtained videos. The proposed algorithm assists the physical therapists in achieving a quantitative method for monitoring the evolution of the patient's posture and allows periodic reviews to be made more quickly, accurately and throughout the physiotherapeutic treatment.