手部追蹤旨在預測影像序列中多個手的軌跡，對於空中手寫、手語辨識及手勢辨識等應用具有重要的意義，而將雙手分組可以使上述應用實現更複雜的功能。 本論文提出基於YOLOv3和聯合嵌入的方法，整合多目標追蹤和關節點檢測的單階段類神經網路模型和演算法，實現實時的多人雙手追蹤。;Hand tracking aims to predict the trajectory of multiple hands in an image sequence, which is of great significance for applications such as air handwriting, sign language recognition and gesture recognition, and grouping the hands can enable the above applications to achieve more complex functions. This paper proposes a single-stage neural network model and algorithm based on YOLOv3 and associative embedding, integrating multi-target tracking and joint point detection, to achieve real-time multi-person hand tracking.
