Rapid identification and cleanup of road debris are essential for enhancing
traffic safety and ensuring unobstructed road conditions. Traditional detection
methods often face challenges in accurately identifying debris in complex
environments with varying light and weather conditions. To address these
limitations, this study proposes a deep learning-based road debris detection
method designed for improved accuracy and robustness. First, road images are
processed using a semantic segmentation approach to remove background
information, isolating only the drivable areas. This segmented region is then
subjected to further object detection to filter out typical non-debris objects,
such as vehicles, pedestrians, and non-motorized vehicles, thereby retaining a
focused image that only contains potential debris or spill objects. Lastly, the
processed image is compared to a baseline image to detect differences and
identify road debris with high precision. Through these steps, the proposed
method achieves accurate detection of road debris even in challenging
environmental conditions. Experimental results validate the model’s
effectiveness, indicating that it provides a reliable solution for debris
recognition, ultimately offering robust technical support for intelligent
traffic management systems aimed at promoting safer, cleaner, and more efficient
roads.