Human Activity Recognition
This repository performs Human Activity Recognition, where given a video, the developed framework recognizes the activity of the object per-frame using a pre-Trained 3D convolutional ResNet-34 model. The applications of it includes:-
- Automatically classifying/categorizing a dataset of videos on disk.
- Verifying that a food service worker has washed their hands after visiting the restroom or handling food that could cause cross-contamination (i.e,. chicken and salmonella).
- Monitoring bar/restaurant patrons and ensuring they are not over-served.
Requirements
NumPyOpenCVcollections
Usage
Data
- The pre-Trained 3D ResNet-34 model of the developed framework is trained on the Kinetics dataset.
- It consists of:-
- 400 human activity recognition classes
- At least 400 video clips per class (downloaded via YouTube)
- A total of 300,000 videos
- The full list of classes the model can recognize can be seen here.
Training and Testing
- The pre-Trained weights of the 3D convolutional ResNet-34 model, trained on the Kinetics dataset, can be downloaded from here.
- To see the implementation of Human Activity Recognition without rolling prediction, check
Human_Activity_Recognition.py. Moreover, to test this implementation on a test video, run the following command:-python Human_Activity_Recognition.py --model resnet-34_kinetics.onnx --classes Class_Labels/action_recognition_kinetics.txt --input example_activities.mp4 - To see the implementation of Human Activity Recognition with rolling prediction, check
Human_Activity_Recognition_Deque.py. Besides, run the following command to check the performance of this implementation:-python Human_Activity_Recognition_Deque.py --model resnet-34_kinetics.onnx --classes Class_Labels/action_recognition_kinetics.txt --input example_activities.mp4
Results
Below are some results of developed model on test videos:-
| Test Video 1 | Test Video 2 | Test Video 3 |
|---|---|---|
 |
 |
 |
| Test Video 4 | Test Video 5 |
|---|---|
 |
 |