摘要
针对立体视觉原理的新型立体折反射全向成像系统结构设计和面向立体柱面全景像对的局域灰度相关对应点快速匹配算法,从捕获的全向立体影像中提取有效深度信息,用于辅助全向视频分析处理中的对象检测和跟踪。采用单相机和两个不同参数的抛物面型反射镜构造了一种共轴结构的折反射全向立体成像装置,捕获的存在一定视差的原始全向立体像对被投影展开为立体柱面全景像对,而后通过特定对应点匹配算法提取稠密的深度信息。对应点匹配算法采用局部区域灰度相关的算子,并充分利用了双向匹配和柱面全景的外极线约束来提高匹配的速度和准确度。仿真实验有效恢复了场景深度信息,证明了整套装置结构设计及深度估计方法的有效性。
A novel structure of Stereo catadioptric omni - directional imaging system and the corresponding points match method for the stereo cylindrical panorama pairs are discussed in order to effectively estimate the depth information in the captured stereo omni -directional images, which will be helpful to the object detection and tracing in the omni - directional video analysis and processing. The structure of a coaxial catadioptric omni - directional stereo imaging system is designed using one camera and two different parabolic reflectors. The captured original stereo omni - directional image pairs with definite parallax to each other are transformed into the cylindrical panorama pairs, and then dense depth information is extracted using the given corresponding points match method. The corresponding points match based on the local area grayness related function makes full use of bidirectional match and the epipolar line constraint of the cylindrical panorama pairs to improve the match efficiency. The simulated experiment restores the scene depth correctly, which shows the validity of the designed system configuration and the local area grayness related corresponding points match method.
出处
《计算机仿真》
CSCD
2008年第12期246-250,共5页
Computer Simulation
基金
国家自然科学基金(No.60773023)
国家自然科学基金(No.60705013
国家自然基金资助(No.60872150))资助
关键词
立体视觉
折反射全向成像
柱面全景
深度估算
对应点匹配
Stereo vision
Catadioptric omni- directional imaging
Cylindrical panorama
Depth estimation
Corresponding points match
作者简介
王炜(1973-),男(汉族),陕西宝鸡,博士,副教授,研究方向为多媒体与虚拟现实。
熊志辉(1976-),男(汉族),江西南昌,博士,讲师,研究方向为数字图像处理,全景成像及嵌入式系统技术。
徐玮(1973-),男(汉族),江西南昌,博士,讲师,研究方向为虚拟现实技术。
崇洋(1981-),男(汉族),辽宁沈阳,硕士研究生,研究方向为全景成像技术。、