運用自動影像分析進行動態跟隨及跌倒偵測的系統 ; A System for Dynamic Tracking and Fall Detection Using Automated Image Analysis
Saved in:
| Title: | 運用自動影像分析進行動態跟隨及跌倒偵測的系統 ; A System for Dynamic Tracking and Fall Detection Using Automated Image Analysis |
|---|---|
| Authors: | 呂承泰, LU, CHENG-TAI |
| Contributors: | 資訊管理學系 |
| Publication Year: | 2025 |
| Collection: | Chinese Culture University: Institutional Repository (CCUR) |
| Subject Terms: | 智慧型居家照護, 跌倒辨識, 人臉辨識, 即時追蹤, OpenPose, Smart home care, Fall detection, Facial recognition, Real-time tracking |
| Description: | 隨著許多國家都面臨人口快速老年化，人力照護的供給遠遠趕不上需求的增加，使得居家照護成為近年相當熱門的研究議題，因此遠端監控系統以及AI影像的分析技術陸續產生許多相關的研究成果，但過去這類居家照護的相關研究有三大主要缺點，首先、透過定點式攝影設備總會有相當程度的影像擷取的死角，並且也無法跟隨被監控照護者的移動路徑，以致無法隨時掌握被照護者最新的情況。其次、過去的照護監控系統也無法有效分析與辨識被照護者的真實跌倒情況，如此便容易讓老人延誤被立即適當的得到醫療處理的時機。第三、過去研究無法做初步判斷老人家跌倒的嚴重程度，但是這點對於通報時狀況的能否有較清楚的描述，對於緊急的醫療處理其實會有非常關鍵的作用，因此如能在跌倒的第一時間、便能透過即時影像與分析做出更加準確的跌倒嚴重程度判讀的重要參考，將是非常關鍵與重要的。 因此、本研究運用無人機的機動性配合Dlib HOG演算法進行人臉的辨識，以達到即時追蹤與跟隨被照護者的目的，其次、本研究針對OpenPose即時多人動作分析模組進行改良與提升對於跌倒的各種分析判斷能力，以開發出能夠正確分析出被照護者跌倒當下的大致真實情況，例如像摔倒的方向性與姿勢的判讀。最後、本研究將進一步針對被照護者跌倒的持續時間，進行嚴重程度的進一步判斷，然後再傳給預設好的通報對象去進行急救和後續醫療的處理。因此、本研究將進行相關的跌倒影像的分析與判斷的實驗，以檢視本研究摘要所描述過去相關研究的三大缺點與不足之處，預期將能更有效地提升居家照護的即時追蹤性，以及及時跌倒的狀況判斷與給予照護的正確性，同時、亦期望能達成更有效且準確地提供跌倒的通報內容，以達到幫助許多孤單老人的智慧型居家照護的安全性與有效照護性。 As many countries face rapid population aging, the demand for caregiving far exceeds the available supply, making home care a popular research topic in recent years. Consequently, remote monitoring systems and AI-based image analysis technologies have emerged, producing numerous relevant research outcomes. However, previous studies on home care have three major limitations. First, fixed-point camera systems inevitably have blind spots in image capture and cannot follow the movements of the monitored individuals, making it challenging to consistently track their real-time status. Second, traditional caregiving monitoring systems often fail to accurately analyze and identify actual fall incidents, which may delay elderly individuals from receiving timely and appropriate medical attention. Third, past research has been unable to preliminarily assess the severity of falls, which is crucial for providing clear and detailed information during emergency reporting. The instant and accurate information about the severity of falls will support the subsequent related medical judgment. In this study, we leverage the mobility of drones combined with the Dlib HOG algorithm for facial recognition to achieve real-time tracking and following of the individuals being monitored. Furthermore, we enhance the OpenPose ... |
| Document Type: | thesis |
| Language: | Chinese |
| Relation: | https://irlib.pccu.edu.tw//handle/987654321/54541; https://irlib.pccu.edu.tw/bitstream/987654321/54541/2/index.html |
| Availability: | https://irlib.pccu.edu.tw//handle/987654321/54541 https://irlib.pccu.edu.tw/bitstream/987654321/54541/2/index.html |
| Accession Number: | edsbas.C0AC90A5 |
| Database: | BASE |
Nájsť tento článok vo Web of Science | Abstract: | 隨著許多國家都面臨人口快速老年化，人力照護的供給遠遠趕不上需求的增加，使得居家照護成為近年相當熱門的研究議題，因此遠端監控系統以及AI影像的分析技術陸續產生許多相關的研究成果，但過去這類居家照護的相關研究有三大主要缺點，首先、透過定點式攝影設備總會有相當程度的影像擷取的死角，並且也無法跟隨被監控照護者的移動路徑，以致無法隨時掌握被照護者最新的情況。其次、過去的照護監控系統也無法有效分析與辨識被照護者的真實跌倒情況，如此便容易讓老人延誤被立即適當的得到醫療處理的時機。第三、過去研究無法做初步判斷老人家跌倒的嚴重程度，但是這點對於通報時狀況的能否有較清楚的描述，對於緊急的醫療處理其實會有非常關鍵的作用，因此如能在跌倒的第一時間、便能透過即時影像與分析做出更加準確的跌倒嚴重程度判讀的重要參考，將是非常關鍵與重要的。 因此、本研究運用無人機的機動性配合Dlib HOG演算法進行人臉的辨識，以達到即時追蹤與跟隨被照護者的目的，其次、本研究針對OpenPose即時多人動作分析模組進行改良與提升對於跌倒的各種分析判斷能力，以開發出能夠正確分析出被照護者跌倒當下的大致真實情況，例如像摔倒的方向性與姿勢的判讀。最後、本研究將進一步針對被照護者跌倒的持續時間，進行嚴重程度的進一步判斷，然後再傳給預設好的通報對象去進行急救和後續醫療的處理。因此、本研究將進行相關的跌倒影像的分析與判斷的實驗，以檢視本研究摘要所描述過去相關研究的三大缺點與不足之處，預期將能更有效地提升居家照護的即時追蹤性，以及及時跌倒的狀況判斷與給予照護的正確性，同時、亦期望能達成更有效且準確地提供跌倒的通報內容，以達到幫助許多孤單老人的智慧型居家照護的安全性與有效照護性。 As many countries face rapid population aging, the demand for caregiving far exceeds the available supply, making home care a popular research topic in recent years. Consequently, remote monitoring systems and AI-based image analysis technologies have emerged, producing numerous relevant research outcomes. However, previous studies on home care have three major limitations. First, fixed-point camera systems inevitably have blind spots in image capture and cannot follow the movements of the monitored individuals, making it challenging to consistently track their real-time status. Second, traditional caregiving monitoring systems often fail to accurately analyze and identify actual fall incidents, which may delay elderly individuals from receiving timely and appropriate medical attention. Third, past research has been unable to preliminarily assess the severity of falls, which is crucial for providing clear and detailed information during emergency reporting. The instant and accurate information about the severity of falls will support the subsequent related medical judgment. In this study, we leverage the mobility of drones combined with the Dlib HOG algorithm for facial recognition to achieve real-time tracking and following of the individuals being monitored. Furthermore, we enhance the OpenPose ... |
|---|