Added Face Mask Detection Deep Learning Model using YOLOv7

[saleena-18]

# Related Issues or Bug

• Info about Issue or Bug: The current implementation of the face mask detection system was using outdated models that do not perform well in varying environmental conditions or on different face types. Issues have been observed with false positives and negatives, reducing the overall accuracy of the system.

• Fixes: #1787 This PR addresses issue #1787 , which is related to the need for improving the accuracy and robustness of the face mask detection model.

# Proposed Changes

• Info about Changes: This pull request introduces the YOLOv7 model into the face mask detection project. YOLOv7 has been selected due to its superior performance in real-time object detection tasks compared to previous models. Key changes include:
  • Integration of YOLOv7 for improved detection in accuracy and speed.
  • All the comparison shown using the hyperparameters like the F1 curve, P curve, R curve, PR curve, confusion matrix.
  • Used preprocessing and postprocessing techniques in Roboflow to optimize the model's performance.

# Additional Info

• Anything Related to Issues: YOLOv7 has demonstrated significant improvements over other models such as YOLOv5 and Faster R-CNN, especially in terms of detection speed and accuracy. Previous models struggled with high false-positive rates in cluttered environments and with varying mask types. YOLOv7's advanced architecture and feature extraction capabilities address these issues effectively. This project benefits from YOLOv7's state-of-the-art performance, providing a more reliable and efficient solution for real-time face mask detection.

# Screenshots

Explanation of YOLOv7's Advantages

• YOLOv7 outperforms its predecessors by offering higher accuracy and faster inference times, which is crucial for real-time applications.
• YOLOv7's advanced feature extraction and architectural enhancements significantly reduce false positives and negatives.

Issue with Implementing Other Models

• YOLOv5: While YOLOv5 offers good performance, it has been shown to be less effective in high-density scenarios and may suffer from lower accuracy in complex environments.
• Faster R-CNN: Known for its accuracy, but it is relatively slower in detection speed compared to YOLO models, making it less suitable for real-time applications.