Foggy weather conditions significantly impact the visibility required for autonomous vehicle navigation, resulting in performance degradation of on-road object detectors. Mobile computing devices deployed at the edge are often resource-constrained and cannot cope well with the performance degradation of detectors. Image dehazing and ensuring the resource efficiency of dehazing algorithms are crucial for any navigation task of vehicle on the road. To ensure safe and smooth operation in foggy weather conditions, the clarity of the image needs to be improved. To address the challenges, an improved detection model called Lightweight Defog Detector (LDD) is proposed to improve detection performance. First, the dark channel prior and positional normalization algorithm is used to dehaze the image and improve image clarity. Second, a lightweight MOELAN feature extraction module is constructed, significantly improving detection efficiency and making it suitable for deployment on edge devices. Finally, the attention mechanism is introduced to further extract features from the feature maps used for detection, which improves the model's ability to extract feature information. Experiments are conducted on the foggy on-Road dataset. The experimental results show that compared with the YOLOv9 and other models, such as IA-YOLO, the proposed algorithm has a slightly higher average accuracy and can effectively improve the performance of vehicle detection in foggy weather while reducing the parameters.