Research on real-time detection of fabric defects based on an improved Elo rating algorithm

To further enhance the detection efficiency of complex textured fabric defects and meet real-time detection demands, an improved Elo rating algorithm is proposed. Initially, the traditional Elo rating algorithm is refined, with a focus on threshold calculation and detection processes. The concept of integral images is incorporated to reduce computational demands and improve operational speed. The refined algorithm is then applied to detect three types of fabric samples, achieving an overall detection accuracy exceeding 80%. Parameter analysis indicates that selecting a sub-region count R between 10 and 30 yields optimal results, with larger sub-regions demonstrating higher detection accuracy. Comparative experiments with five object detection models are conducted, introducing two evaluation metrics: mean Average Precision (mAP) and Frames Per Second (FPS). Results show the algorithm outperforms others in mAP and FPS (102.1), thereby meeting industrial real-time detection requirements.