Question about LSTM frame window size and other ML approaches

[YoungjaeDev]

Hi,

Nice project! I have a couple of questions:

1. What frame window size did you use for the LSTM model? (e.g., 30 frames, 60 frames, etc.)

2. Did you try any tabular-based ML approaches like Random Forest or Lightgbm with engineered features from the keypoints?

Thanks!

[Ian-I-Chien]

Hi, nice to meet you

1. I’m using a 120-frame window (60 fps, about two seconds). The two-second choice isn’t from a specific study — just my assumption that a full fall usually takes that long. This can be adjusted later.
2. I haven’t tried tabular methods. Otherwise, my YOLO+LSTM idea is still tentative and may not be the best, so I’d be glad to explore your suggestions too.