In complex urban traffic environments where standard YOLOv8n architectures struggle with limited feature fusion and optimal localization, we propose a new CN-YOLOv8n model. This architecture integrates a Bi-directional Feature Pyramid Network (BiFPN) to enhance multi-scale feature aggregation, thereby suppressing background-induced false positives and increasing classification confidence. Recognizing that BiFPN tends to weaken small object and low resolution feature representations, an improved attention mechanism, EffectiveSE, is introduced to reinforce feature expressiveness and substantially improve the detection of small or occluded objects. To refine localization accuracy without increasing model complexity, we introduce the MPDIoU loss function, which jointly considers spatial distance and dimensional disparity between predicted and ground truth bounding boxes. Extensive experiments on the self collected vehicle and pedestrian dataset demonstrate that CN-YOLOv8n achieves significant gains in detection accuracy and robustness, particularly for small or heavily occluded targets, underscoring its suitability for real world intelligent transportation applications.