Intrackability Detection Based Upper-Limb Motion Recognition

Aging of population becomes a widely concerned social issue in the coming decades. To provide proper services (e.g., smart home) for the elderly, it is expected to accurately recognize their daily activities without wearing devices, especially upper-limb motions. Thus, this paper proposes a contactless upper-limb motion recognition method based on intrackability motion detection and sliding window of BOW (IMD-SWBOW) model which can recognize single-person's motions and interactive motions via camera videos in daily scenes. The intrackability-based detection algorithm is adopted to improve the accuracy of STIPs extraction. SWBOW model and the nearest neighbor classification method are applied. An SD-Upper-Limb motion video database is established for experiments which includes 10 kinds of continuous single-person and interactive upper-limb motions performing by 19 subjects in daily scenes and dynamic background (such as chaotic scenes, camera shake or zoom and different outfits). The experimental results show that the average recognition accuracy is 91%. The proposed method can improve the recognition performance of upper-limb interactive motions in daily monitoring scenes with dynamic background.