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

Abstract Falls and gait imbalance are a major health threat to older adults, yet public skeleton datasets that targetelderly adults in real indoor care environments, particularly for clinically critical unbalanced gaits such ascrutch-assisted walking, remain scarce. Here we present ElderlyPipe, a dynamic skeleton dataset of elderlyabnormal gait recorded in real indoor care environments. The release comprises 63,855 quality-assured skele-ton frames grouped into 287 continuous segments and 28,735 standardized 24-frame windows, covering fourbehaviours: standing, walking, crutch-assisted walking (using_crutch), and sitting. The skeletons are com-pleted by hierarchical occlusion compensation and refined by visibility-aware adaptive Kalman smoothing.Benchmarks on the release show that the temporal association of skeleton sequences underpins imbalanceanalysis: a spatiotemporal graph model outperforms its single-frame counterpart in overall F1 (0.656 vs.0.605) and better separates crutch-assisted walking, the representative unbalanced state, from everyday be-haviours. The anonymized skeleton sequences, metadata, cross-validation splits, documentation, the completeprocessing pipeline, and the baseline code are openly released under a persistent identifier.