Action Recognition using YOLO and Feature Extraction

This repository contains an implementation for action recognition using the YOLO object detection algorithm, and feature extraction using either OSNet or face recognition.

Dependencies

• scikit-learn
• PIL
• ultralytics
• numpy
• opencv-python
• argparse
• time

How to Use

1. Clone the repository
2. Install the required dependencies
3. Run the main script, main.py, with your desired arguments.

Example Usage:

python main.py -method 'osnet' -detection_confidence 0.8 -num_saved_images 30 -verification_time 2 -threshold_coefficient 0.8 -input_video 'test_videos/test_video3.mp4' 

Arguments
-method: method to use for feature extraction (default: 'osnet')
-detection_confidence: detection confidence threshold (default: 0.75)
-num_saved_images: number of saved images for the target person (default: 30)
-verification_time: time interval between verifications (default: 3)
-threshold_coefficient: coefficient for similarity threshold (default: 0.9)
-save_frames: save frames with the title of similarity score and yes/no for verification (default: False)
-input_video: input video file

### Results_analysis.ipynb
This notebook contains the results of the experiments conducted on the dataset. It also contains the code for the plots and tables in the report.

### Blog Post
See the blog post here: https://medium.com/@justasand1/project-title-bf3dff25a791