Wiki » History » Version 8
Version 7 (Petri Partanen, 19/05/2020 17:32) → Version 8/10 (Petri Partanen, 19/06/2020 20:33)
h1. Running instructions Wiki
h2. Promising
# Download `project_gait.zip` https://github.com/marian-margeta/gait-recognition
```
$ python dummy_pose_estimation.py
Traceback (most recent call last):
File "dummy_pose_estimation.py", line 5, in <module>
from "Files" in Redmine. (https://redmine.fe.up.pt/attachments/download/21825/gait_project.zip) scipy.misc import imresize, imread
# Extract ImportError: cannot import name 'imresize'
```
Solved by using ImageIO:
```
import imageio
from PIL import Image
from human_pose_nn import HumanPoseIRNetwork
net_pose = HumanPoseIRNetwork()
net_pose.restore('models/MPII+LSP.ckpt')
img = imageio.imread('images/patient.png')
img = Image.fromarray(img).resize((299, 299))
```
h2. Meeting notes
h3. 19.5.2020
Important:
Look into optical flows! There's speed patterns which cluster and thus can be used to classify gait.
Discussion:
In the downloaded file project report we have to be able to answer the following questions:
# Open terminal * state why the particular method was selected
* what we compared to
* perfomance in the extracted folder literature
# Run * we have to know what we are talking about. for example, in case of neural network cnn and the following commands whole architecture
If using machine learning:
* consider different metrics/parameters to perform the abnormal and normal gait classification
* study which parameters yield the best classification results
If using the neural network to build the skeleton:
* from skeleton, extract features, then use classical machine learning (knn, svm, etc) to perform the classification.
Questions:
* Is there a specific programming language that perform better in terminal this domain? No, we can choose any programming language. Matlab and Python might be the most commonly used.
* If we were to choose the neural network to skeleton approach, how could we extract gait parameters and features from the skeelton? The shared research paper provides inspiration how to extract typical bioimechanical gait features.
h2. Evaluated software
h3. DLTdv8
# Feature extraction: `$python main.py` MATLAB program to perform motion analysis. Promising but automatic point tracking didn't work. Neural network point tracker feature of the software was not evaluated.
# Classifier: `$python classify.py` https://bitbucket.org/thedrick/dltdv/src/default/
h3. Mokka - Motion kinematic & kinetic analyzer
Professional software for motion analysis. Requires C3D files, thus not applicable to our project.
h3. MOtoNMS
https://github.com/RehabEngGroup/MOtoNMS
h2. To be evaluated:
https://www.pyimagesearch.com/2015/05/25/basic-motion-detection-and-tracking-with-python-and-opencv/
https://github.com/gsimchoni/mocap
https://github.com/gsimchoni/mocap/blob/master/R/readAMC.R
https://github.com/browarsoftware/RMoCap
h2. Promising
# Download `project_gait.zip` https://github.com/marian-margeta/gait-recognition
```
$ python dummy_pose_estimation.py
Traceback (most recent call last):
File "dummy_pose_estimation.py", line 5, in <module>
from "Files" in Redmine. (https://redmine.fe.up.pt/attachments/download/21825/gait_project.zip) scipy.misc import imresize, imread
# Extract ImportError: cannot import name 'imresize'
```
Solved by using ImageIO:
```
import imageio
from PIL import Image
from human_pose_nn import HumanPoseIRNetwork
net_pose = HumanPoseIRNetwork()
net_pose.restore('models/MPII+LSP.ckpt')
img = imageio.imread('images/patient.png')
img = Image.fromarray(img).resize((299, 299))
```
h2. Meeting notes
h3. 19.5.2020
Important:
Look into optical flows! There's speed patterns which cluster and thus can be used to classify gait.
Discussion:
In the downloaded file project report we have to be able to answer the following questions:
# Open terminal * state why the particular method was selected
* what we compared to
* perfomance in the extracted folder literature
# Run * we have to know what we are talking about. for example, in case of neural network cnn and the following commands whole architecture
If using machine learning:
* consider different metrics/parameters to perform the abnormal and normal gait classification
* study which parameters yield the best classification results
If using the neural network to build the skeleton:
* from skeleton, extract features, then use classical machine learning (knn, svm, etc) to perform the classification.
Questions:
* Is there a specific programming language that perform better in terminal this domain? No, we can choose any programming language. Matlab and Python might be the most commonly used.
* If we were to choose the neural network to skeleton approach, how could we extract gait parameters and features from the skeelton? The shared research paper provides inspiration how to extract typical bioimechanical gait features.
h2. Evaluated software
h3. DLTdv8
# Feature extraction: `$python main.py` MATLAB program to perform motion analysis. Promising but automatic point tracking didn't work. Neural network point tracker feature of the software was not evaluated.
# Classifier: `$python classify.py` https://bitbucket.org/thedrick/dltdv/src/default/
h3. Mokka - Motion kinematic & kinetic analyzer
Professional software for motion analysis. Requires C3D files, thus not applicable to our project.
h3. MOtoNMS
https://github.com/RehabEngGroup/MOtoNMS
h2. To be evaluated:
https://www.pyimagesearch.com/2015/05/25/basic-motion-detection-and-tracking-with-python-and-opencv/
https://github.com/gsimchoni/mocap
https://github.com/gsimchoni/mocap/blob/master/R/readAMC.R
https://github.com/browarsoftware/RMoCap