Initial Ventures with OpenCV
My journey began with basic models using OpenCV. I started by developing systems that could detect faces but quickly discovered they were easily deceived by hands or other objects. This stage exposed the limitations in accurately identifying mask usage and underscored the need for more sophisticated solutions. Despite these early challenges, these projects provided essential insights into the complexities of face detection and set the foundation for future advancements.
Advancing with TensorFlow and Improved Detection
To achieve better accuracy, I moved on to using TensorFlow for more robust mask detection. By training models on larger datasets and applying data augmentation techniques, I improved the system's ability to distinguish masks from other objects. However, I noticed that these models still had difficulties with identifying masks that were not properly worn. This phase marked a significant improvement in my work and set the stage for further enhancements.
Utilizing Advanced Architectures: MobileNet, ResNet, and YOLOv5
Determined to overcome previous limitations, I integrated advanced architectures like MobileNet V2 and ResNet into my projects to enhance detailed detection of masks. These models allowed the system to accurately detect whether a mask was fully covering the face, including the nose, reducing false negatives and increasing reliability. Building on these advancements, I adopted YOLOv5 for real-time mask detection, known for its speed and accuracy. With YOLOv5, I could quickly recognize various mask-wearing states such as fully covered faces, masks pulled down, and the absence of masks, demonstrating my commitment to using cutting-edge technology for public health.
