Design of Fall Detection System based on YOLOv3
Lingli Chen, Shunkai Zhang, Gang Li, Haojie Wang · 2023
Aiming to address the risk of falls among the elderly, it proposes a human fall detection system based on YOLOv3. To improve the convergence and detection performance of the original YOLOv3 network, a loss function based on Intersection Over Union is employed for improvement. With the same detection time, the improved YOLOv3 network achieved a detection accuracy of 98.3%, which is 1.3% higher than before the improvement. After training completion, the pt format model was converted into the wk format that can be recognized by the development board and ultimately transplanted onto Hisilicon Hi3559AV100 device to run, completing the hardware implementation of the fall detection system. Testing showed an accuracy rate of 93.5% in different scenarios.