Object detection in low-light urban environments presents significant challenges due to poor illumination, low contrast, and blurred object boundaries. Conventional real-time detectors, such as YOLOv8, often experience performance degradation under these conditions, resulting in missed detections and inaccurate localization. In this work, we propose Edge-Aware YOLOv8 (EA-YOLOv8), a novel extension of the YOLOv8 architecture that integrates an edge-detection branch to enhance contour sensitivity and spatial awareness. By fusing gradient-based edge features with semantic representations at multiple stages of the network, our model improves boundary-level precision without compromising inference speed. We validate EA-YOLOv8 on two benchmark datasets, ExDark and BDD100K-night, and report substantial gains over the baseline YOLOv8, achieving up to a +6.5% improvement in mAP@0.5:0.95. Extensive ablation studies confirm the effectiveness of fusion depth, edge detector choice, and fusion weight sensitivity. Additional analyses, including gradient norm tracking, Grad-CAM visualizations, and a comprehensive accuracy–latency–power trade-off, demonstrate that EA-YOLOv8 is both interpretable and efficient, making it well-suited for real-time deployment on edge devices. The proposed method provides a lightweight yet powerful solution for enhancing object detection robustness in safety-critical, low-light environments, such as autonomous driving and smart urban surveillance.