摘要
为了进一步增加均值平移算法用于目标跟踪的准确性和鲁棒性,提出了加入目标位置的预测、对大目标间断采样、联合颜色信息与边缘信息建立直方图、采用多个候选目标优选与验证等4种改进方法,减小算法的计算量,提高算法的匹配速度,增强算法的适应性。使用Edinburgh大学提供的Workshop 1 front图像序列进行测试,分别得到了原算法与改进算法的Bhattacharyya距离,通过比较其平均值和标准偏差,说明了改进算法中最优匹配比原算法准确。使用自拍摄的复杂背景图像测试后,说明了改进算法的适应性。实验结果表明,改进后的均值平移算法对运动目标跟踪效果良好。
To improve the precision and robustness of Mean-shift algorithm which is used to track the moving target, four methods were put forward to increase the matching speed, improve the adaptability of the algorithm and decrease the calculation workload. The four methods were importing position prediction for target, discontinuous-sampling for large target, combing color and edge information of target to compute histogram, choose the optimal result from several target candidates. The Bhattacharyya distance of the original and improved algorithm was obtained by using Workshop 1 front image sequence provided by Edinburgh University. Comparing the mean and standard mean-square error of Bhattacharyya distance, the optimal matching of improved algorithm was more accurate than the original algorithm. The test result show that the adaptability of improved algorithm is good by using image under complex background photographed. The improved Mean-Shift algorithm is effective to track moving target.
出处
《红外与激光工程》
EI
CSCD
北大核心
2008年第3期556-560,共5页
Infrared and Laser Engineering
基金
总装备部试验技术研究项目(2003SY4106007)
关键词
目标跟踪
均值平移算法
直方图
Tracking object
Mean-Shift algorithm
Histogram
