Research on Infrared Target Detection Technology Applied to Nighttime Scenarios

During nighttime driving, it is common to encounter situations where the headlights of oncoming vehicles are too bright, posing a challenge to the driver's vision and reaction time, thereby increasing the risk of traffic accidents. Infrared target detection technology, by capturing the thermal radiation of vehicles and pedestrians, provides an effective method to improve nighttime driving vision. In this paper, a new lightweight deep learning object detection network, YOLO-GC, is proposed. First, Constructing the necessary dataset for this study has addressed the issue of limited infrared datasets, providing a solid data foundation for the training and evaluation of the model. Second, the backbone model of YOLOv5, CSPDarknet53, was replaced with the GhostNet network, reducing the computational load and model size while maintaining high accuracy at lower computational complexity. Finally, an attention mechanism was introduced to focus on key areas of the image, specifically on regions with significant temperature changes in infrared images, enhancing the accuracy of object detection. Experimental results show that the improved network achieves a recognition accuracy of 94.2% for vehicles and pedestrians, an improvement of 3.9% compared to the original network.