Infrared and visible image fusion (IVIF) aims to create fused images that encompass the comprehensive features of both input images, thereby facilitating downstream vision tasks. However, existing methods often overlook illumination conditions in low-light environments, resulting in fused images where targets lack prominence. To address these shortcomings, we introduce the Illumination-Aware Infrared and Visible Image Fusion Network, abbreviated by IAIFNet. Within our framework, an illumination enhancement network initially estimates the incident illumination maps of input images, based on which the textural details of input images under low-light conditions are enhanced specifically. Subsequently, an image fusion network adeptly merges the salient features of illumination-enhanced infrared and visible images to produce a fusion image of superior visual quality. Our network incorporates a Salient Target Aware Module (STAM) and an Adaptive Differential Fusion Module (ADFM) to respectively enhance gradient and contrast with sensitivity to brightness. Extensive experimental results validate the superiority of our method over seven state-of-the-art approaches for fusing infrared and visible images on the public LLVIP dataset. Additionally, the lightweight design of our framework enables highly efficient fusion of infrared and visible images. Finally, evaluation results on the downstream multi-object detection task demonstrate the significant performance boost our method provides for detecting objects in low-light environments.

With the development of science and technology, artificial intelligence has been widely used in the transportation field, and research on the symmetry of artificial intelligence has become increasingly more in-depth. Traffic sign detection based on deep learning has the problems of different target shapes and high variability in the number of targets between different labels. To solve these problems from a lack of symmetry, the idea of applying the concept of balanced data and the deformable positioning region to a target recognition network is proposed. The research is based on the improvement of the Libra R-CNN. Aiming at the problem that the difficult-to-distinguish target in target detection has a high impact on detection, the idea of generating increasingly more diverse indistinguishable samples during training is proposed to improve the detection accuracy, which is verified by experiments. The experiment is carried out on the MS COCO 2017 and traffic sign datasets. The improved Libra R-CNN is 3 percentage points better than the unimproved Libra R-CNN’s mean Average Precision (mAP). A large number of comparative experimental results show that the improved network is effective.

Multi-Object Tracking (MOT) detects multiple objects in videos and assigns unique identities across consecutive frames, generating complete motion tracks. Existing methods typically associate fixed-size predicted tracks with detections above a threshold, but objects with complex motion patterns, such as sudden changes in speed or direction, often cause motion estimation errors, leading to missed associations and fragmented tracks. To tackle this, we analyze the impact of attributes like speed, size, and detection confidence on tracking performance. Based on this, we propose an adaptive scaling strategy that dynamically adjusts the size of predicted tracks to create adaptively scaled bounding boxes. These bounding boxes enable a Motion-Adaptive IoU (MA-IoU) association strategy for more robust matching between detections and predicted tracks. Additionally, we introduce a self-supervised training method to optimize the adaptive scaling network and MA-IoU across diverse scenarios. Finally, we cascade MA-IoU to develop SMART (Scalable Motion-Adaptive Robust Tracking), a robust and efficient multi-object tracker. SMART achieves state-of-the-art results on BEE24 and DanceTrack. On the BEE24 benchmark, SMART achieves 67.2 MOTA, 63.4 IDF1, and 47.9 HOTA, demonstrating superior performance in handling objects with complex motion and indistinguishable appearances. This comprehensive approach addresses the limitations of prior methods, offering a significant advancement in multi-object tracking.

Multi-object tracking (MOT) in low-light environments presents significant real-world application value. Despite substantial progress in MOT, low-light MOT remains constrained by the scarcity of specialized datasets, largely because collecting and manually annotating low-light tracking data is difficult and often prohibitively expensive. This paper specifically addresses these challenges through methodological and dataset innovations. We first present low-light multi-object tracking (LLMOT), the first comprehensive low-light MOT dataset containing 11,580 images, including 5316 labeled and 6264 unlabeled images, consisting of nighttime-enhanced MOT17 sequences and multiple unannotated low-light videos. To simultaneously alleviate the constraint of annotation costs and address the damage that low-light-induced image degradation causes to pseudo-label quality, we propose Consistency Regularization Track (CRTrack), a semi-supervised framework tailored for low-light scenarios. Specifically, we introduce a consistent adaptive sampling assignment mechanism that calibrates and filters noisy and shifted pseudo-bounding boxes under low-illumination conditions. We then design an adaptive semi-supervised network update strategy that enables the model to more stably exploit unlabeled low-light videos for iterative optimization. Extensive experiments on the LLMOT dataset validate the effectiveness and robustness of the proposed method. CRTrack achieves 62.472 HOTA, 71.544 MOTA, and 75.864 IDF1 on the LLMOT dataset, demonstrating its effectiveness in low-light MOT. Our approach provides a practical solution for low-light MOT tasks with significant real-world implications.