I want an initial direction, like how can I use IMU sensors on knees (e.g upper leg and lower leg ) to estimate body state e.g running, walking, climbing, etc. The basic idea I have is to use 2 IMU sensors on knee position (upper and lower leg) , and to get data or make a data set of it. Then process it using deep learning e.g CNN or ANN etc. But point is..are there data sets available on which I can test CNN etc to see either it works or not. Need guidance about data sets, from where could I get IMU knee based data sets, so that I can focus on my algorithm only.
I believe the question should be the "action" of the knee, not the state. I think if you have two IMU sensors, then you will be able to construct the pose of the knee wrt a common frame (body, hip). As the knee joint is a 1DOF joint, you might want to fuse with the speed of the frames for more information. So probably the state space of the combined frame would look like {joint angle, angular speed, linear speed}. As long as your tasks are mutually exclusive {climbing stairs, walking, running} you might be able to use a simple classifier like KNN. But I'm not quite sure where can you get an exact set of data for your dynamic model.
It seems you are trying to solve a classic problem of human activity recognition based on body-worn sensors. A very good starting point for different datasets, mainly based on acceleration only, is this paper: Article UniMiB SHAR: a new dataset for human activity recognition us...
I wouldnt recommend using a naked CNN for testing. A classic machine learning approach with a proper preprocessing, some good extracted features (e.g. simpe statistical features, PCA or Codebook) and a classification that supports the understanding of the problem (e.g. kNN or SVM). Good luck and have fun trying :).
- Badges
- Science topic