Design of a Multi-Object Wearable Recognition and Tracking Algorithm Based on PHSM-YOLO

This project focuses on the design and development of a multi-object wearable recognition system using advanced deep learning-based object detection and classification techniques. The system is developed to detect and classify important safety-related wearable objects and human presence, including Helmet, Vest, Person, bare-arms, Gloves, Non-Helmet, and Shoes. The main objective of this project is to improve workplace and industrial safety monitoring by automatically identifying whether individuals are wearing required protective equipment. For effective detection, two advanced models are implemented: YOLOv8+BiFPN+SimAM+DyHead and RT-DETR+CST+CBAM. The YOLOv8-based model is enhanced with BiFPN for better multi-scale feature fusion, SimAM attention for improving important feature representation, and DyHead for strengthening object localization and classification. The RT-DETR-based model integrates CST and CBAM modules to improve feature extraction, attention learning, and detection accuracy in complex environments. The project includes both classification and object detection tasks, enabling the system to identify multiple wearable classes from input images with improved accuracy and reliability. A user-friendly application is designed to allow users to upload images and obtain detection results with bounding boxes and class labels. This system supports automated safety compliance checking and reduces the need for manual observation. The proposed approach provides an efficient, accurate, and intelligent solution for multi-object wearable recognition in industrial, construction, and workplace safety environments.

Keywords: Multi-Object Detection, Wearable Recognition, YOLOv8, RT-DETR, BiFPN, SimAM, DyHead, CBAM.