基于嵌入式CPU-GPU的高清鱼眼视频实时校正系统

摘要

In the field of security video surveillance, real-time monitoring system is needed to achieve real-time monitoring for 360°×180° wide panoramic areas with high definition, existing fish-eye correction systems have higher cost, poor flexi-bility, especially low definition and poor real-time problems, etc. In order to solve real-time problems, this article proposes CPU-GPU high-speed communication protocol based on embedded platform STiH418 and CPU-GPU memory sharing method based on programming shader, and achieves HD panoramic fisheye video real-time correction system on CPU-GPU using texture mapping technology. Compares with the related correction system, experimental results show that the system takes good algorithm efficiency, effectiveness and integrity of corrected image, can fully meet the HD pano-ramic 360° × 180°(4 million pixels, 2048 × 2048p30)in real-time correction, embedded systems STiH418 reduce overall system cost compared with PC, correction MAP is generated and updated by ARM CPU in software, virtual PTZ improves system flexibility and stability, so the system has a high practical value in security video surveillance market.%在安防监控领域，需要鱼眼实时监控系统实现360°×180°大范围高质量无死角全景实时监控，现有的鱼眼校正系统存在成本较高，灵活性差，特别是清晰度不高和实时性差等方面的问题。针对如何提高全景高清鱼眼视频校正的实时性问题，提出了基于嵌入式平台STiH418的CPU-GPU高速通信协议和基于可编程着色器的嵌入式CPU-GPU内存共享方法，并利用GPU的纹理映射技术实现了全景高清鱼眼视频实时校正系统。实验结果表明，与相关校正系统相比，该系统很好地兼顾到算法效率、图像校正效果和完整性，可以完全满足360°×180°的全景高清（400万像素，2048×2048p30）鱼眼视频实时监控，而且与使用PC服务器相比嵌入式系统降低了系统整体成本， ARM CPU软件生成更新校正算法和可事时实时和事后的虚拟PTZ提高系统灵活性和稳定性，因此该系统具有很高的实用价值。