车载前向运动视频的实时全景成像方法

发布时间：2018-05-04 11:52

本文选题：机器视觉 + 全景成像　；参考：《光学学报》2017年05期

【摘要】：传统的视频拼接模型和图像对齐方法中,耗时的图像匹配和复杂的光流计算一直是实时拼接的性能瓶颈。提出了一种车载前向运动视频的实时全景成像方法,该方法充分利用物理场景的几何结构和列车的运动信息来构造视频的拼接区域,并建立几何模型实现了拼接区域的对齐。整个拼接避免了复杂的图像处理过程,实现了铁路全景图的实时获取,同时也提供了一种轻量级的视频全景索引方式,降低了视频数据的存储和访问开销。
[Abstract]:In the traditional video mosaic models and image alignment methods, the time-consuming image matching and complex optical flow calculation have been the performance bottleneck of real-time stitching. This paper presents a real-time panoramic imaging method for vehicle forward moving video. The method makes full use of the geometric structure of physical scene and the motion information of train to construct the mosaic region of the video, and establishes a geometric model to realize the alignment of the stitched region. The whole mosaic avoids the complex image processing process and realizes the real-time acquisition of railway panoramic images. It also provides a lightweight video panoramic index method which reduces the storage and access cost of video data.
【作者单位】：北京交通大学交通数据分析与挖掘北京市重点实验室;中国铁道科学研究院基础设施检测研究所;
【基金】：国家973计划(2013CB329400) 国家自然科学基金(61272354,61273364,61300176,61473031,61472029) 北京市自然科学基金(4152042)
【分类号】：TP391.41

【相似文献】