摘要： 危险区域人员监控具有摄像节点密度高，传输实时性和图像质量要求高的特点。功率域非正交多址接入（Power Domain Non-Orthogonal Multiple Access，PD-NOMA）技术可以支持多路并行传输，有利于在密集传输场景下提升传输实时性，而多摄像节点协同有利于提高图像质量。该文研究在面向危险区域人员监控的PD-NOMA摄像网络中如何通过多图融合实现高质量监控。首先，定义了图像的“单人信息量”这一关键术语，它反映了图像中单个人员被准确识别的概率。其次，基于摄像节点之间的位置关系定义了多摄像节点拍摄到的单个人员的融合图像信息量。最后，在满足传输实时性要求以及图像中所有人员均可识别的前提下，以最大化融合图像信息量为目标，计算以摄像节点的图像分辨率和无线发射功率为控制变量的实时传输调度方案。实验评估表明，当实时传输上限为0.4秒时，基于PD-NOMA的传输调度方案传输信息量比传统传输方案提高了46.4%，使得图像中人员识别概率从0.8549提升至0.8919。随着实时传输上限值的放松，识别概率增长率随之快速下降。
Abstract: Personnel monitoring in hazardous areas requires high camera density, real-time transmission, and high image quality. Power Domain Non-Orthogonal Multiple Access (PD-NOMA) technology can support the parallel transmission of multiple channels, which is beneficial for improving real-time transmission in dense transmission scenarios. Furthermore, the collaboration of multiple camera nodes can improve image quality. This paper investigates how to achieve high-quality monitoring in PD-NOMA camera networks for personnel monitoring in hazardous areas through multi-image fusion. The term "single-person information" is defined as a key concept that reflects the probability of accurately identifying a single person in an image. Then, the fusion image information of a single person captured by multiple camera nodes is defined based on the spatial relationship between the camera nodes. To maximize the fusion image information, a real-time transmission scheduling scheme is calculated with the camera node's image resolution and wireless transmission power as control variables, while satisfying the real-time transmission requirements and the recognition of all personnel in the image. Experimental evaluations show that when the real-time transmission upper limit is 0.4 seconds, the transmission information of the PD-NOMA-based transmission scheduling scheme is 46.4% higher than that of the traditional transmission scheme, increasing the probability of personnel identification in the image from 0.8549 to 0.8919. As the real-time transmission upper limit is further relaxed, the growth rate of the identification probability rapidly decreases.