Implement basic activity-detection

[pescador9]

To achieve activity-detection we want to detect the activity of people by using ML models.

First find a model which detects poses, then collect pose-data for different activities and try different models.

Acceptance Criteria:

• create a first working model for activity-detection
• A set of activities is defined which can be detected (sitting, standing)
• progress is documented for reports (Wiki + Slides)

[egeatmaca]

TODO:

Week 1:

• [x] 1. Choose preliminary pre-trained pose detection model.
• [x] 2. Implement pose detection with the preliminary pre-trained model.
• [x] 3. Collect data for activity classification from pose. Labels: Sitting laying back and working. Collect data until Wed, May 18 that:
• resembles the real office setup with table and chair
• shows the person in standing or sitting positions
• [x] 4. Find meaningful features for the activity classification from pose model. First ideas: Elbow angle, back and leg angle

  Week 2:

• [x] 5. Prepare features
• [x] 6. Train, validate and test the classification model.
• [x] 7. Embed the classification model into code and test the whole concept with a real time camera.

[egeatmaca]

Links

Links for Todo 1:

1. https://viso.ai/deep-learning/pose-estimation-ultimate-overview/
2. https://www.linkedin.com/pulse/top-9-pose-estimation-models-2022-ritesh-kanjee/

Links for Todo 3:

1. https://tele-immersion.citris-uc.org/berkeley_mhad

[egeatmaca]

Feature Ideas for Task 4:

Angles:

Neck
Shoulder
Elbow
Hip
Knee
Between calf and floor