Safety Threat Detection in Construction Zones using Dense YOLOv8 with White Shark Nested Attention Network

• Main Authors: J, Rajeeth T, Dalai, Chitaranjan, Naik, Vanitha G, Vekariya, Vipul, B, Kannadasan, Patil, Harshal
• Format: Conference Proceeding
• Language: English
• Subjects: Accuracy; Construction safety; Data models; Feature extraction; Grid-Constrained Data Cleansing Method; Noise; Object detection; Real-time systems; Safety; Safety management; Sharks; Threat assessment; White Shark Nested Attention Network; YOLOv8
• ISSN: 2768-0673
• DOI: 10.1109/I-SMAC61858.2024.10714823

The construction industry faces numerous hazards and risks, many of which remain unmonitored, leading to accidents and injuries. To address these challenges and improve safety in construction areas, this paper proposes a novel approach based on the Dense YOLOv8 White Shark Nested Attention Network (DYWS-NAN) for safety threat detection. Utilizing the Construction Site Safety Image Dataset, which contains a wide range of images depicting safety compliance and violations, the proposed method ensures precise identification of safety threats. The dataset undergoes rigorous pre-processing using the Grid-Constrained Data Cleansing Method to remove noise and enhance image quality, enabling the model to focus on critical safety features. Following pre-processing, feature extraction and classification are performed using the DYWS-NAN architecture, which combines YOLOv8's real-time object detection capabilities with the White Shark Nested Attention mechanism. This integration enhances both object feature extraction and classification performance, significantly improving safety threat detection accuracy. Simulations conducted in the Python environment demonstrate an accuracy level of 99.4%, highlighting the method's effectiveness in reducing false alarms and enabling timely safety interventions in construction zones. This approach enhances safety management and ensures better compliance with safety regulations.