CAJ | 학술논문

使用多次曝光法进行高动态范围视频获取时,需要将多幅低动态范围(LDR)图像进行快速融合,并满足视频融合系统的实时性要求.本文提出了一种改进的快速多分辨率塔形分解融合算法,在不降低原有算法融合质量的前提下极大减少计算量.算法分别从金字塔分解卷积核、高斯系数扩展插值方法、拉普拉斯系数融合规则、高斯系数融合规则等方面进行改进,简化了塔形分解、融合和重构过程.该方法应用在某运载火箭视频采集系统中,实验结果表明,该改进算法稳定性强,融合质量高,并极大地减少了计算量,运算速度提高了10倍以上,在 TMS320DM642 DSP 上一帧512×512图像的融合时间约为8.9 ms,满足25 fps 高动态范围视频融合的实时性要求.
사용다차폭광법진행고동태범위시빈획취시,수요장다폭저동태범위(LDR)도상진행쾌속융합,병만족시빈융합계통적실시성요구.본문제출료일충개진적쾌속다분변솔탑형분해융합산법,재불강저원유산법융합질량적전제하겁대감소계산량.산법분별종금자탑분해권적핵、고사계수확전삽치방법、랍보랍사계수융합규칙、고사계수융합규칙등방면진행개진,간화료탑형분해、융합화중구과정.해방법응용재모운재화전시빈채집계통중,실험결과표명,해개진산법은정성강,융합질량고,병겁대지감소료계산량,운산속도제고료10배이상,재 TMS320DM642 DSP 상일정512×512도상적융합시간약위8.9 ms,만족25 fps 고동태범위시빈융합적실시성요구.
Fast fusion of multiple low dynamic range (LDR)images is essential for the real-time per-formance of a high dynamic range (HDR)video acquisition system using multi-exposure method.This paper proposed an improved fusion algorithm based on multi-resolution pyramid decompose.The algo-rithm dramatically reduces computing complexity while preserving the fusion quality of the original method.The proposed method improves the real-time performance in the following aspects:pyramid decompose convolution core,Gaussian coefficients expansion method,Laplacian coefficients fusion rule,Gaussian coefficients fusion rule.The computing complexity of pyramid decomposes,fusion, and reconstruction is reduced.Experiment results shows that the proposed method is stable and got good fusion results.The method is applied to a launch vehicle observation system.The computing speed is boosted 10 times and it takes 8.9 ms to fuse one 5 12×5 12 frame on TM320MD642 DSP,ful-filling the real-time requirement of 25 fps HDR video acquisition system.