Dates: 2022
Principal Investigator: Dr. Hamzah Luqman
Description: PPE violation detection is important to enhance workers' safety in the workplace. Automating the process of PPE monitoring helps in ensuring the worker's compliance and reducing hazards and threats on site. PPE monitoring is one of the computer vision problems that involve several phases for detecting PPE compliance in video streams. The proposed system will consist of four main stages: detection, segmentation, features learning, and classification. In the first stages, the worker will be detected in the video stream. Further detection and segmentation will be performed based on the target PPE. For example, the system will detect the worker's head to monitor helmet PPE violation.  Then, feature learning will be devised for the PPE and used as input to the machine learning classifier. Since having sufficiently large instances in the dataset that are labeled is not an easy task, we extend the work to build models that employ unsupervised feature learning as well. We will evaluate several deep learning techniques. In addition, we will develop an augmented database of PPE using the site camera. The collected database will be used in training and evaluating the proposed system.