This work addresses the fundamental challenge of object detection in low-quality visual environments, where poor illumination, sensor noise, motion blur, and compression artifacts together degrade the quality of features and reduce the accuracy of detection. To alleviate these limitations, we propose a unified architecture called Hybrid Noise-Aware Vision Transformer (HN-ViT) that simultaneously enhances image quality and improves detection robustness. The framework is composed of a Noise-Aware Enhancement module and a dual-stream feature extraction backbone that fuses convolutional neural networks with Vision Transformers. In this regard, the proposed NAE module significantly enhances the visual quality, boosting +9.7 dB in PSNR and + 0.27 in SSIM for degraded inputs. Extensive experimental evaluation on benchmarks such as ExDark, SIDD, and COCO-Lite indicates significant improvements in the detection performance by HN-ViT with IoU improvements in a range of 8–13% and mAP improvements of 6–10% compared to the state-of-the-art CNN and transformer-based detectors like YOLOv7 and Swin-T. Qualitative results further reveal that HN-ViT facilitates more stable bounding-box localization and a reduction of missed detections under extremely low-light and noise-intensive conditions. Overall, the proposed HN-ViT framework offers an effective, scalable, and noise-resilient solution for real-world degraded imaging environments and thus holds great potential for applications in surveillance, autonomous systems, and edge-based vision.