Infrared object detection via feature interaction and attention-guided fusion

As a key technology of the environmental perception part of the autonomous driving system, the object detection method must accurately locate and recognize traffic objects in real time. However, it often exhibits false positive (FP) or false negative (FN) errors in complex road scenes. The infrared imaging system can clearly image in low-light conditions, making it suitable for object detection in complex environments. However, conventional object detection methods are challenging to extract a robust feature representation from the limited semantic information of small objects, due to the small proportion of pixels occupied by infrared occluded and small objects, as well as the low contrast of infrared images. Therefore, utilizing active and passive infrared cameras to quickly and accurately detect infrared occluded objects and small objects is a challenging task. Aiming at the problem that the object detection algorithm is not effective in detecting infrared occluded objects and small objects in complex road scenes, the improved object detection method in complex infrared scenes (ODMCIS)-you only look once (YOLO) object detection method is proposed, which is based on the network model of you only look once version 8s (YOLOv8s). First, the dual-branch (DB)-spatial pyramid pooling fast (SPPF) module and the dual-residual branch (DRB)-C2f module were designed to enhance and fuse multi-scale features. Then, a loss function was proposed to accelerate the model's convergence speed and improve detection accuracy. Finally, the YOLOv8n network model was improved by proposing a cross-level fusion mechanism, and the detection head was redesigned to make object localization more accurate. The experimental results show that the ODMCIS-YOLO algorithm achieves high precision, and the reasoning speed reaches 137.6 frames per second (FPS), which meets the requirements of real-time detection and lays the foundation for the realization of all-day real-time autonomous driving, and its comprehensive performance is better than that of state-of-the-art (SOTA) object detection method, which can more efficiently complete the task of traffic object detection in complex road scenes.