基于区域分割与融合的全景稳像算法  

Panoramic image stabilization algorithm based on region segmentation and fusion

作　　者：尹丽华[1,2,3] 李范鸣(指导) 刘士建[1] Yin Lihua;Li Fanming;Liu Shijian(CAS Key Laboratory of Infrared System Detection and Imaging Technology,Shanghai Institute of Technical Physics of the Chinese Academy of Sciences,Shanghai 200083,China;University of Chinese Academy of Sciences,Beijing 100049,China;ShanghaiTech University,Shanghai 200031,China)

机构地区：[1]中国科学院上海技术物理研究所,中国科学院红外探测与成像技术重点实验室,上海200083 [2]中国科学院大学,北京100049 [3]上海科技大学,上海200031

出　　处：《红外与激光工程》2018年第9期380-389,共10页Infrared and Laser Engineering

基　　金：国家863计划(2011AA7031002G);中国科学院上海技术物理研究所创新基金(CX-122)。

摘　　要：车载红外全景扫描成像系统具有每列单独成像、360°全方位视场覆盖的特点,从而导致传统的电子稳像算法无法直接适用,因此,提出一种基于区域分割与融合的全景稳像算法。首先,通过局部列偏移调整方法对图像预补偿。接着,以车头前进方向为基准,对全景图像进行区域分割,即前端、右端、后端、左端区域。然后,根据各区域的成像特点,选择不同的稳像模型进行稳像,其中,运动估计环节采用滑窗策略缩短运算时间,运动补偿环节采用未定义区重构方法弥补边界缺失信息。最后,利用局部区域扩展、渐入渐出加权平均融合方式对重叠区域进行区域拼接、融合,保证全景图像无缝拼接。实验结果表明:该算法有效解决了车体行进过程中红外全景扫描系统的稳像问题,稳像关键指标——帧间峰值信噪比(PSNR)可以提高14.7%,运行时间可缩短为传统算法的1/10,基本满足了工程应用的需求。Vehicle infrared panoramic scanning imaging system has the characteristics of column single imaging and 360°full field view,which result in that traditional electronic image stabilization methods could not be directly applied,therefore,the panoramic image stabilization algorithm based on region segmentation and fusion was proposed.First,the image pre-compensation was completed by local column offset adjustment method.Then the panoramic image was divided by the area,i.e.,the front region,the right region,the end region and the left region,based on vehicle forward direction.Then each region used different image stabilization models for stabilization according to imaging characteristics,and the sliding window strategy was used to shorten the computation time in motion estimation process,the undefined area reconstruction method was used to compensate for missing information in motion compensation process as well.Finally,the local area expansion strategy and the progressive fade-weighted average fusion strategy were adopted to complete the region′s splicing and fusion,resulting in the seamless splicing effect of panoramic images.Experimental results indicate that the proposed method which can meet the requirements of engineering applications can effectively solve image stabilization problems of the system in the course of the vehicle moving forward,the precision of Peak Signal to Noise Ratio(PSNR)is increased by 14.7%compared with the original image sequence,running time is reduced to the 1/10 of traditional algorithms.

关 键 词：全景扫描成像 电子稳像 分割 融合 帧间峰值信噪比 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]