一种小重叠区红外序列图像快速拼接算法

A fast stitching algorithm for infrared sequence images of small overlap region

鉴于全景图像拼接所需的重叠比例大,拼接时间长,提出了一种小重叠区单目红外全景图像的快速拼接算法。首先,待拼接的红外图像序列通过搭载于云台的红外图像探测器绕轴定点均匀拍摄采集得到,之后在感兴趣区域(region of interest,ROI)内使用改进加速的小重叠区红外图像配准算法得到相邻图像的配准信息,确定相邻图像的重叠区域,然后统计重叠区域的亮度信息进行亮度差异调节以获得整体图像的视觉一致性,亮度调整后的重叠区域图像通过使用三角函数权重的方式进行图像融合,最后对剩余的待拼接图像重复上述步骤,得到过渡自然、视觉一致性良好的超宽视野红外全景图。在Ambarella平台上的实测结果表明,该方法能够成功拼接重叠区域较小(6%)的红外图像,拼接速度快,实时性高,具有良好的应用前景和实用价值。

In view of the large overlap ratio and the long stitching time required for panoramic image,a fast stitching algorithm for monocular infrared panoramic images with small overlap region is proposed.Firstly,the infrared image sequence to be stitched is collected by the infrared image detector mounted on pan-tilt by shooting uniformly around the axis.Then,the improved and accelerated small overlap region infrared image registration algorithm is used in the region of interest(ROI)to obtain registration information of the adjacent images and determine the overlap area of adjacent images.Next,the brightness information of the overlap area is calculated and the brightness difference is adjusted to obtain the visual consistency of the overall image,and image fusion of the overlap area images after brightness adjustment is performed by using trigonometric function weights.Finally,the above steps are repeated for the remaining images to be stitched to obtain an ultra-wide-field infrared panoramic image with natural transition and good visual consistency.The actual measurement results on the Ambarella platform show that the method can successfully stitch infrared images with a small overlap area(6%),with high stitching speed and high real-time performance,and has good application prospects and practical value.