Yongsheng Qiu

Although Transformer-based real-time object detectors have achieved promising performance in autonomous driving scenarios, their ability to detect small objects remains limited. This limitation primarily arises because small objects occupy only a few pixels in an image and contain weak edge and texture information, which can be further degraded during feature extraction and multiscale feature propagation. To address these issues, this study proposes a wavelet-based frequency-aware feature enhancement method using RT-DETR as the baseline network. First, a Wavelet Frequency Unit is introduced into the feature fusion stage of the RT-DETR neck. The unit employs the Haar wavelet transform to decompose the input features into low- and high-frequency subbands, thereby decoupling information across different frequency components. Second, residual enhancement and a frequency attention mechanism are applied to strengthen edge and texture details in the high-frequency branch. Finally, the low-frequency subband is fused with low-resolution features across scales, followed by feature reconstruction using the inverse wavelet transform. This design improves the representation of small objects in the feature space. Tests on KITTI and BDD100K verify the method. On KITTI, it obtains 95.5% mAP@0.5 and 69.7% mAP@0.5:0.95, exceeding the RT-DETR baseline by 1.8 and 1.1 percentage points. APs and ARs rise by 2.7 and 2.3 percentage points. On the selected BDD100K subset, the corresponding mAP@0.5 and mAP@0.5:0.95 values are 51.7% and 29.7%.

In rainy and foggy conditions, the scattering of light and the occlusion effects of atmospheric particles distort the reflected light from object surfaces, leading to inconsistent depth information. As a result, depth estimation models trained under clear weather conditions fail to generalize effectively to adverse weather conditions. To address this challenge, we propose a novel CNN-Transformer architecture, WeatherMono, for self-supervised monocular depth estimation under rainy and foggy weather. Rainy and foggy images often contain large regions of low contrast and blurry features. By combining Convolutional Neural Networks (CNNs) with Transformers, WeatherMono effectively captures both local and global contextual information, thus improving depth estimation accuracy. Specifically, we introduce a Multi-Scale Deformable Convolution (MDC) module and a Global-Local Feature Interaction (GLFI) module. The MDC module extracts detailed local features in rainy and foggy environments, while the GLFI module incorporates an efficient multi-head attention mechanism into the Transformer encoder, enabling more effective capture of both local and global information. This enhances the model’s ability to comprehend image features, strengthens its capability to handle low-contrast and blurry images, and ultimately improves the accuracy of depth estimation in adverse weather conditions. Experiments on WeatherKITTI show WeatherMono achieves AbsRel of 0.097, outperforming WeatherDepth (0.104) and RoboDepth (0.107). On DrivingStereo, it achieves AbsRel of 0.149 (rain) and 0.101 (fog). Extensive qualitative and quantitative experiments demonstrate that WeatherMono significantly outperforms existing methods in terms of both accuracy and robustness under rainy and foggy conditions.

Image dehazing is a challenging ill-posed problem in low-level computer vision tasks, requiring the restoration of high-quality, haze-free images from complex and foggy conditions. Deep learning-based dehazing methods struggle to effectively remove non-homogeneous fog distributions due to the uneven and dense nature of fog patches, making it difficult to clear real-world fog variations. A key challenge for non-homogeneous image dehazing algorithms is efficiently capturing the spatial distribution of haze in areas with varying fog densities while restoring fine image details. To address these challenges, we propose MLCANet, a multi-level composite attention-guided network for non-homogeneous image dehazing. MLCANet mitigates the impact of uneven haze areas through two main components: the Multi-level Composite Attention Generation Network (MCAGN) and the Dehazed Image Reconstruction Network (DIRN). The MCAGN integrates channel attention (CA), spatial attention (SA), and multi-scale pixel attention (MSPA) to capture haze features at different spatial scales. The DIRN, based on a decoder-encoder architecture, combines multi-scale dilated convolutions and deformable convolutions to restore fine image details more flexibly and efficiently. Extensive qualitative and quantitative experiments, along with ablation studies, demonstrate the effectiveness and feasibility of this method for non-homogeneous image dehazing.

Convolutional neural network (CNN)-based autonomous driving object detection algorithms have excellent detection results on conventional datasets, but the detector performance can be severely degraded in low-light foggy weather environments. Existing methods have difficulty in achieving a balance between low-light image enhancement and object detection. To alleviate this problem, this paper proposes a foggy traffic environment object detection framework, IDOD-YOLOV7. This network is based on joint optimal learning of image defogging module IDOD (AOD + SAIP) and YOLOV7 detection modules. Specifically, for low-light foggy images, we propose to improve the image quality by joint optimization of image defogging (AOD) and image enhancement (SAIP), where the parameters of the SAIP module are predicted by a miniature CNN network and the AOD module performs image defogging by optimizing the atmospheric scattering model. The experimental results show that the IDOD module not only improves the image defogging quality for low-light fog images but also achieves better results in objective evaluation indexes such as PSNR and SSIM. The IDOD and YOLOV7 learn jointly in an end-to-end manner so that object detection can be performed while image enhancement is executed in a weakly supervised manner. Finally, a low-light fogged traffic image dataset (FTOD) was built by physical fogging in order to solve the domain transfer problem. The training of IDOD-YOLOV7 network by a real dataset (FTOD) improves the robustness of the model. We performed various experiments to visually and quantitatively compare our method with several state-of-the-art methods to demonstrate its superiority over the others. The IDOD-YOLOV7 algorithm not only suppresses the artifacts of low-light fog images and improves the visual effect of images but also improves the perception of autonomous driving in low-light foggy environments.