LIVDN: low illumination vehicle detection network

Detecting vehicles under low illumination conditions poses a significant challenge due to reduced visibility and lack of contrast. To address this issue, this paper proposes a Low Illumination Vehicle Detection Network (LIVDN). LIVDN utilizes the Dilation-Wise Residual module to enhance the feature extraction network, allowing for a more comprehensive capture of contextual information. The Bidirectional Cascade Feature Fusion module improves detection capabilities for vehicles of various sizes. Additionally, a Bi-level Routing Spatial Attention module directs the network’s attention to vehicle texture features and color information, enhancing detection accuracy. The proposed method is validated on the BDD100K dataset and KITTI dataset. Experimental results demonstrate a significant improvement in vehicle detection accuracy under low illumination conditions.