摘要
监控摄像机的隔行扫描是造成视频降质的重要因素,因此需要在进一步处理前对其进行去隔行。针对典型的视频监控场景,提出了一种快速有效的运动自适应去隔行算法。采用同极性4场运动检测法提取运动信息,将图像分为静止区域、运动区域和混合区域;对静止区域采用直接的场合并,对于运动区域采用改进的基于边缘的插值,混合区域则采用基于运动向量的加权平均。为提高算法的实时性,基于Nvid ia的CUDA技术对算法进行了GPU加速。实验结果表明,本文提出的去隔行算法插值精度高,边缘处理效果好,经过GPU加速后,处理速度也得到了大幅提高。
The interlaced scan of surveillance cameras is the main defect for degrading video quality. Thus deinterlacing is needed to convert interlaced video into progressive scan format for future processing. We propose a fast and effect motion-adaptive deinterlaeing algorithm for typical video surveillance application. The algorithm employs same-parity 4-filed motion detection to extract motion information. Then the image is divided into three parts： static area, motion area and mixed area. Field merging is applied to static area directly. In the motion area, an improved edge-based interpolation is implemented carefully. And in the mixed area, a weighted average on motion vector is used. To speed up the algorithm, NVIDIA CUDA technology is utilized. Experimental results show that the proposed method achieves high-precision interpolation and provides good results in edge regions, and the processing time is decreased significantly by GPU acceleration.
出处
《传感技术学报》
CAS
CSCD
北大核心
2010年第3期393-398,共6页
Chinese Journal of Sensors and Actuators
