Building a Dataset of Pointing Gestures for Elderly People in Iraqi Nursing Homes

Elderly and stroke patients with speech and hearing difficulties often struggle with communication, particularly in complex environments. This paper presents a new dataset specifically created for semantic segmentation, instance segmentation, and sign recognition to meet the needs of these individuals. The dataset pertains to elderly Iraqis and was collected from several nursing homes. Approval from the relevant authorities has been obtained. The images in the dataset have been pre-processed, enhanced, and annotated to facilitate effective model training and evaluation. By applying the YOLOv8 model for gesture detection and classification, we achieved precision, recall, and mAP50 scores of 95.6%, 93.9%, and 97.1%, respectively. These findings highlight the dataset's potential to improve assistive technologies for better communication support.