Abstract: An SVM (support vector machine) classifier is used to classify the pixels and generate a reliable target probability distribution image The outstanding trust region numeric optimization algorithm is utilized to locate the target as well as determine its size The SVM classifier's low error rate of classification offers a better base for the trust region algorithm Comparison with existing method and present tracking examples show that the proposed method is better in both target detection and precision improvement

Key words: support vector machine, scale-space theory, trust region algorithm

摘要： 通过使用SVM(支持矢量机)分类器对像素分类进行目标检测，将输入图像转换成可靠的目标概率分布图，然后结合使用性能优良的信任域优化算法，在概率分布图上实现目标定位并确定其尺寸。分类器对像素分类的低错分率为信任域算法提供了更好的基础。通过和现有算法的比较以及在真实序列图像上的实验表明，该算法不但能够更准确地检测到目标，而且在跟踪精度方面有了显著提高。

关键词: 支持矢量机, 尺度空间理论, 信任域算法