A high-performance framework for personal protective equipment detection on the offshore drilling platform

Abstract

In order for the offshore drilling platform to operate properly, workers need to perform regular maintenance on the platform equipment, but the complex working environment exposes workers to hazards. During inspection and maintenance, the use of personal protective equipment (PPE) such as helmets and workwear can effectively reduce the probability of worker injuries. Existing PPE detection methods are mostly for construction sites and only detect whether helmets are worn or not. This paper proposes a high-precision and high-speed PPE detection method for the offshore drilling platform based on object detection and classification. As a first step, we develop a modified YOLOv4 (named RFA-YOLO)-based object detection model for improving localization and recognition for people, helmets, and workwear. On the basis of the class and coordinates of the object detection output, this paper proposes a method for constructing position features based on the object bounding box to obtain feature vectors characterizing the relative offsets between objects. Then, the classifier is obtained by training a dataset consisting of position features through a random forest algorithm, with parameter optimization. As a final step, the PPE detection is achieved by analyzing the information output from the classifier through an inference mechanism. To evaluate the proposed method, we construct the offshore drilling platform dataset (ODPD) and conduct comparative experiments with other methods. The experimental results show that the method in this paper achieves 13 FPS as well as 93.1% accuracy. Compared to other state-of-the-art models, the proposed PPE detection method performs better on ODPD. The method in this paper can rapidly and accurately identify workers who are not wearing helmets or workwear on the offshore drilling platform, and an intelligent video surveillance system based on this model has been implemented.