-
A Flexible Fall Detection Framework Based on Object Detection and Motion Analysis
First detects human objects in the frame using YOLOv5, then tracks the object by using YOLOv5 every frame.
-
Improved Pedestrian Fall Detection Model Based on YOLOv5Doesn't use motion analysis, but uses YOLOv5 and a CNN to detect pose.
- Detect people in the frame
YOLOV5- Tensorflow pose detection (movenet)
- For each estimated pose:
- calculate the angle of the head and the hip
- calclulate the rotation speed
- calculate the angular acceleration
- calculate the position of the hip
- calculate the difference between y velocity and acceleration
- calculate the width and height ratio of the bounding box
- calculate the angle of the head and the hip
- If the angular acceleration of the head and the hip is greater than a threshold, and the difference between y velocity and acceleration is greater than a threshold, and the width > height, then the person is falling.