结合图像增强和恢复的增强现实跟踪注册

增强现实中的跟踪注册技术一直是研究的重点和难点，而地下矿道和巷道内亮度低，产生的图像较昏暗，车载相机快速运动和抖动，对传统的基于特征匹配的跟踪注册提出了挑战。本文从提高增强现实中跟踪注册稳健性和精度出发，采用基于Retinex改进的方法增强昏暗图像的亮度，同时利用基于对抗神经网络的方法恢复运动模糊图像。首先提取图像ORB特征，实现初始化；然后根据跟踪特征点的数量，开启图像增强和图像恢复线程，提高特征点提取质量和数量。在数据集和真实模拟场景下的试验结果显示，跟踪精度提高了12%左右，在低亮度和含有轻微模糊的情况下，跟踪注册稳健性也有显著提高。

Augmented reality tracking registration combined with image enhancement and recovery

Tracking registration technology in augmented reality has always been the focus and difficulty of research. However, the lowlight of underground mines and roadways resulting relatively dark image,and the on-board camera moving quickly and jitter,which challenge the traditional tracking registration based on feature matching. Starting from improving the robustness and accuracy of tracking registration in augmented reality, an improved method based on Retinex is used to enhance the brightness of dim images, and an adversarial neural network-based method is used to restore motion-blurred images. First, image ORB features is extracted to achieve initialization, and then the image enhancement and image restoration thread are started according to the number of tracked feature points to improve the quality and number of feature point extractions. Experimental results in the data sets and real simulation scenarios show that the tracking accuracy is improved by about 12%, and the robustness of tracking registration is also significantly improved in the case of low brightness and slight blur.