We tried to solve the problem of detecting whether people in a given image are wearing a PPE (Personal Protective Equipment) kit and also which parts of the PPE kit have been worn. Since object detection is a problem that has already been well studied, we would like to use transfer learning solve this problem. We plan to use neu-ral networks trained on popular datasets and adapt them to this domain of PPE detection so that we can achievefaster learning with a small amount of data. This problem fits in the larger picture of autonomously identifying whether safety precautions are being followed by medical professionals in different healthcare scenarios. Adherence to safety precautions in a busy medical setting during a global pandemic is difficult to enforce upon all members of the medical staff, so our model would be highly useful in this situation.

In many object detection problems, it happens frequently that we do not have as much data as a high-quality model would ideally desire. We feel our problem is important to solve, especially in scenarios like a pandemic or an outbreak of a rare disease, when there isn’t enough time to collect a large amount of training data. In recent times, machine learning techniques like transfer learning, class-imbalanced learning and few-shot learning have enabled such kind of applications to be developed. We think using these for the problem of PPE detection and in general the domain of healthcare would be highly beneficial.