摘要
针对单景视觉会因目标颜色与背景颜色相同、或者遮挡物等干扰因素影响,容易导致跟踪目标丢失等,提出智能全景视觉传感网络目标跟踪方法,利用抛物面的发射镜与CCD的摄像机组合成折反射全景成像系统,通过两个成像系统拍摄的全景图像相应匹配点和已知的全景成像镜面参数,计算获得两幅图像彼此间的对极几何关系。再利用滤波清除光流噪声、通过二值化分割目标,获得待跟踪目标,接着在某一帧目标周围查询下一帧位置,敏锐感知目标较倾向起始位置周围的较小移动,同时尺度变化则需要令边界框尽可能保持起始位置的附近,完成目标精准跟踪。实验证明,所提方法不会因为外界干扰和背景颜色相同情况导致无法跟踪目标,且网络目标跟踪效率较高。
In view of the fact that single scene vision can easily lead to the loss of tracking targets due to the same target color and background color, or obstructions and other interference factors, an intelligent panoramic vision sensor network target tracking method is proposed.The catadioptric panoramic imaging system was generated by parabolic reflector and CCD camera.According to the corresponding matching points of panoramic images captured by two imaging systems and the known mirror parameters of panoramic images, the antipode geometric relationship between the two images was obtained.Filtering was used to eliminate optical flow noise;Binarization was used to segment the target and obtain the target to be tracked.When the next frame position was queried around the target in one frame, the small movement around the target and the initial position was perceived;Meanwhile, the scale change was kept near the initial position, thus achieving the accurate target tracking.The experiments show that this method can not track the target because of the same external interference and background color, and the network target tracking efficiency is high.
作者
田维飞
夏志丽
TIAN Wei-fei;XIA Zhi-li(North University of China,Shanxi Taiyuan 030051,China)
出处
《计算机仿真》
北大核心
2021年第9期203-206,261,共5页
Computer Simulation
基金
山西省社会科学界联合会2020至2021年度重点课题(SSKLZDKT2020071)。
关键词
智能全景视觉
传感网络
目标跟踪方法
对极几何约束
Intelligent panoramic vision
sensor network
target tracking method
antipode geometric constraints
CCD camera
