Recent researches have shown that object detector is vulnerable to hazy perturbation. To enhance the recognition capability of object detector for such perturbation, this paper proposes a method named IPOD. The The main idea of the proposed method involves incorporating an additional No-Reference Image Quality (NR-IQA) model alongside GCANET to assess the quality of haze-free images. When the input image contains hazy perturbation, the proposed method is applied for noise reduction, and the denoised image’s quality is compared to that of the original input image. If the input image is better than the denoised image, it is directed to object detector. Conversely, if the denoised image exhibits superior quality, it is forwarded to the object detector. To demonstrate the effectiveness of the proposed method, a comprehensive experiment was conducted. In terms of mAP, the proposed method demonstrates remarkable performance, achieving a 92.5% improvement in mAP. This exceeds the figure for GCANET with approximately 89.94%. This results also show the potential ability of the proposed method in practice.