IMUs Based Real-Time Data Completion for Motion Recognition With LSTM
Abstract
and near future to be precisely analyzed. IMUs are position sensors widely used for motion detection. A set of IMUs attached to the body can
measure the past motion data from the subject, but not the current and near future data due to the delay issues. In this paper, I propose a LSTM
model to predict the current and near future IMUs data from the past IMUs data during a walking motion for real-time human motion recognitions. The network contains two LSTM layers with 162 hidden layers. It takes sequence from past, and output predicted timestamps. I used a
pre-collected IMUs dataset to apply on my model and get very accurate performance within 40 predicted timestamps which is enough for a
typical real-time sensor system to solve the delay problem.
Full Text:
PDFReferences
[1] Luo, Y., Coppola, S. M., Dixon, P. C., Li, S., Dennerlein, J. T., & Hu, B. (2020). A database of human gait performance on irregular and
uneven surfaces collected by wearable sensors. Science Data. Published.
[2] C.O.L.A.H. (2015, August 27). Understanding LSTM Networks -- colahs blog. Github, Colahs Blog. https://colah.github.io/posts/2015-
08-Understanding-LSTMs/.
[3] LSTM PyTorch 1.10.0 documentation. (2019). Pytorch. https://pytorch.org/docs/stable/generated/torch.nn.LSTM.html
[4] NAdam PyTorch 1.10.0 documentation. (2019). Pytorch. https://pytorch.org/docs/stable/generated/torch.optim.NAdam.html#torch.
optim.NAdam
[5] torch.optim PyTorch 1.10.0 documentation. (2019). Pytorch. https://pytorch.org/docs/stable/optim.html
[6] torch.utils.data PyTorch 1.10.0 documentation. (2019). Pytorch. https://pytorch.org/docs/stable/data.html
DOI: http://dx.doi.org/10.70711/frim.v3i6.6655
Refbacks
- There are currently no refbacks.
