A real-time industrial safety automation using YOLO architectures leveraging diverse chromatic domains

Object detection is a crucial component of computer vision with extensive industrial applications. It involves identifying and localizing objects within an image or video frame, surpassing simple image classification by providing bounding boxes for detected objects. In this work, authors used a real-time approach for evaluating YOLO object detection models on a welding defect dataset, using a multispectral approach across diverse chromatic domains. Specifically, the performance of YOLOv3, YOLOv5, and YOLOv8 is analyzed using multiple color spaces-RGB, HSV, LAB, and YCbCr. To examine how color representation affects detection outcomes. YOLO's unified architecture optimizes both speed and accuracy, making it well-suited for real-time industrial safety applications. The results indicate that YOLOv8 demonstrates significant improvements in both accuracy and inference speed over previous versions using RGB color space. A normalized mAP@0.5 of 0.592 was achieved using the RGB color space, highlighting YOLOv8's effectiveness in challenging detection tasks under varying visual conditions. The model's lightweight structure and computational efficiency make it an excellent fit for real-time welding defect detection in diverse industrial environments which has been demonstrated using a real-time framework.