Dual-modality Object Detection Approach Utilizing Enhanced and Fused Features
Danni Wang, Zhengci Yang, Chen Ming Fu · 2024
Object detection, as a key research direction in the field of computer vision, has been widely applied in various fields. In this work, we propose a novel deep model, namely the Dual-modality Object Detection Approach Utilizing Enhanced and Fused Features (DODEF) to improve the detection performance of moving targets in airport operational areas. This approach addresses the limitations of single-modality detection models, which often suffer from reduced accuracy in poor lighting conditions. By harnessing both visible-light RGB and thermal infrared imagery, DODEF employs multi-scale dilated convolutions and Efficient Channel Attention (ECA) mechanisms to refine object feature extraction. Also, it further introduces two fusion strategies, including light-perceptionbased adaptive fusion and convolutional-attention-based fusion, to effectively combine dual-modality features. Comprehensive experiments conducted using image data from airport surveillance cameras confirm that DODEF is capable of more sufficiently capturing object features under various lightning conditions when compared to other baseline models, thus yielding better detection performance for object detection on airport surface.