- 學位論文
運用5PKC的骨架分區策略於時空圖卷積網路優化健身動作識別與正確性評估
Using 5PKC-based Skeleton Partition Strategy in Spatio-Temporal Graph Convolution Networks for Optimizing Fitness Action Recognition and Evaluation
摘要
近年來隨著健康意識抬頭，人們對健身需求開始逐漸提高。然而在健身時，如果健身動作不當也會容易對身體造成傷害。加上近幾年COVID-19的肆虐，使得去健身房會增加染疫風險。為此，若能使用自動化人體動作辨識來對使用者的動作持續地進行檢測，將能有效的避免錯誤動作。而使用者在家中也能使用此系統來分析動作，減少出門在外染疫的風險。 人體動作辨識在多媒體計算中已被廣泛的應用，我們希望能透過人體動作辨識來掌握使用者的健身狀況。然而，大多數的人體動作辨識模型多採用CNN來對圖像進行處理，導致容易從背景中引入除人體外不必要的雜訊。 為了解決這個問題，我們以骨架數據作為輸入，並採用時空圖卷積網路（ST-GCN）來對骨架時空間關係進行分析。為了對動作進一步提高準確性，我們還引用了五大動力鏈系統作為分區策略，藉此來探索骨架的分部狀況。最後以健身相關影片來進行訓練和測試，並且與其他人提出的方法進行比較來確認所提出的模型效果。
並列摘要
With the rise of health awareness, people's demand for fitness has gradually increased. Further, nowadays, because of the COVID-19, going to the gym will increase the risk of infection. For this reason, it will be possible to avoid wrong actions if automatic action recognition can detect and judge the human motion of exercises. Users can also use this system to analyze movements at home to reduce the risk of getting infected when they go out. Human body motion recognition has been widely used in multimedia computing. We hope to master the situation of the user's fitness status through human body motion recognition. However, most human action recognition mostly uses CNN-based models to process images, which may introduce unnecessary noise other than the human body from the background. To address this problem, we use the Spatio-Temporal Graph Convolutional Network (ST-GCN) as the backbone and take skeleton data as input to learn skeleton relationships. To further improve the accuracy, we cited The 5 Primary Kinetic Chains as a partition strategy to explore the skeleton partition. Lastly, the videos about fitness are used as training and testing. Also, compared with methods proposed by others to confirm the effect of the proposed model.