ElderlyPipe: A real-world indoor skeleton dataset of elderly abnormal gait including crutch-assisted walking

ElderlyPipe is a skeleton dataset of older adults recorded in real indoor care facilities. It contains 63,855 quality-assured 33-keypoint frames (x, y, z, visibility) covering four behaviours: standing, walking, crutch-assisted walking, and sitting. Frames are organised into 287 continuous segments and 28,735 standardised 24-frame windows (stride 2), and are accompanied by video-stratified 5-fold cross-validation splits. The 249 released clips were captured across five indoor sub-areas summarised under four scene labels: activity rooms and activity areas (6 clips, sufficient lighting, slight occlusion), the nursing-home corridor (217 clips, mixed natural and artificial light, almost no occlusion), the nursing-home lobby (23 clips, backlight, no occlusion), and the public dining area (3 clips, dim lighting, heavy occlusion). Only anonymised numerical coordinates are distributed; no facial imagery is included.