摘要
针对生猪多目标跟踪过程中猪舍内光照情况复杂、生猪间遮挡等问题，改进了传统多运动目标跟踪算法，开发了一套基于计算机视觉技术的目标跟踪系统。改进了前景检测方法，将灰度差分、s通道差分和帧间差分所获取的差分图像相融合。利用阴影与背景的颜色差异性和纹理相似性消除阴影的影响，得到完整的前景图像，提高了前景提取的准确度。改进均值漂移算法实现对目标生猪运动行为的准确跟踪，改进的算法融人了目标生猪的初始位置与颜色信息，修正了目标直方图模型，提高了跟踪的准确率。不同品种、不同光照条件下生猪的运动行为跟踪实验的结果表明，运动目标检测方法能够有效消除阴影对跟踪的影响，同时验证了算法的稳定性，跟踪准确率大于85％。
In order to realize the multi-target tracking of pig＇s behavior in pigsty complex light situations and shelter among pigs, the traditional multi-object tracking algorithm is improved, and a target tracking system based on computer vision technology is developed. The foreground detection algorithms are improved, and the difference images obtained by gray difference, S-channel difference and frame difference are fusion. The color differences and texture similarity between shadow and background are used to eliminate the effect of shadow upon detection. The camshift algorithm is improved to realize the accurate tracking of pig＇s motor behavior in pigsty. The improved algorithm integrates the initial position and color information of target pigs, and fixes the histogram model. It improves the tracking accuracy. By the experiments of tracking different varieties pigs and tracking under different light situations, the experimental results show that the proposed target detection method can effectively eliminate the influence of the shadow and the experiments verify the stability of the algorithm, the tracking accuracy rate is greater than 85 %.
出处
《光学学报》
EI
CAS
CSCD
北大核心
2015年第A01期109-117,共9页
Acta Optica Sinica
基金
北京市科委资助项目（D141100003814003）
关键词
图像处理
猪
多目标跟踪
目标检测
image processing
pig
multi-target tracking
target detection
