In recent years, the innovative development of attention mechanism modules has provided new ideas for algorithm optimization, including large-scale separable kernel attention (LSKA), efficient multi-scale attention (EMA) and dilated multi-scale attention (MSDA). The impact of these attention mechanism modules on the performance improvement of the YOLO model remains to be explored. In this experiment, the Traffic Sign Localization and Detection dataset is used to explore how CBAM can improve the object detection performance of the yolov8 model. Experimental results show that the improved YOLOv8-CBAM model shows significant performance improvements, with a single-frame inference time increase of 0.6 ms, an average accuracy (mAP@50) of 2.1%, and a recall rate of 9.2%. Comparative experiments further reveal that the CBAM module strengthens the feature selection ability through the attention mechanism, especially in complex background or small target detection.