| 顾及行人的室内全景影像拼接方法 |
| |
| 作者姓名： | 董颖青 姚剑 李礼 朱吉 |
| |
| 作者单位： | 武汉大学遥感信息工程学院 |
| |
| 基金项目： | 国家自然科学基金(41571436);国家重点研发计划(2017YFB1302400)。 |
| |
| 摘 要： | 在室内场景的全景影像拼接过程中,易出现行人运动目标引起的“鬼影”现象。针对此问题,提出一种顾及行人的室内全景影像拼接方法,利用深度学习对整幅影像进行逐像素目标级分割,保持行人目标的完整性,并结合重匹配、显著度检测等对运动类型进行判定,生成带权的运动区域掩膜,结合灰度、梯度差及纹理复杂度等特征,融入基于图割算法的拼接线检测能量方程中,对行人区域进行补偿,最终生成背景干净、无“鬼影”的全景影像。
|
| 关 键 词： | 全景影像 目标分割 图割能量优化算法 最优拼接线 |
An Indoor Panoramic Image Stitching Algorithm with Consider of Pedestrian |
| |
| Authors: | DONG Yingqing YAO Jian LI li ZHU Ji |
| |
| Institution: | (School of Remote Sensing and Information Engineering,Wuhan University,Wuhan 430079,China) |
| |
| Abstract: | In the panoramic image mosaic of indoor scene,the problem of ghosts caused by pedestrian is easy to appear.To solve this problem,an indoor panoramic image stitching algorithm with consider of pedestrian is proposed in the paper.Firstly,deep learning is used to obtain a per-pixel instance segmentation result,which could maintain the integrity of the pedestrian.Further,with the pedestrian re-identification and saliency detection,the pedestrian are divided into difference moving types,and the weighted moving area mask is then generated.Finally,the weighted moving area masks with the features of gray,gradient difference and texture complexity are extracted,and putting into the optimal seamline detection in the graph cuts energy minimization framework to generate high-quality panorama with relative clean background and no ghost. |
| |
| Keywords: | panoramic images instance segmentation graph cuts optimal seamline |
| 本文献已被 CNKI 维普 等数据库收录！ |
|
