CAJ | 학술논문

针对分类器学习常常面临高维数据的问题，借助稀疏表示理论对目标样本多尺度Harr特征进行数据降维，构建朴素贝叶斯分类器进行目标正负样本的学习和更新，选择具有最大分类器响应值的样本作为目标的当前状态，实现了对运动目标的快速而有效的跟踪。实验结果表明该方法适用于机器人运动目标跟踪，在提高实时性的同时能保持一定的鲁棒性。
침대분류기학습상상면림고유수거적문제，차조희소표시이론대목표양본다척도Harr특정진행수거강유，구건박소패협사분류기진행목표정부양본적학습화경신，선택구유최대분류기향응치적양본작위목표적당전상태，실현료대운동목표적쾌속이유효적근종。실험결과표명해방법괄용우궤기인운동목표근종，재제고실시성적동시능보지일정적로봉성。
For the problem of high-dimensional data faced with classifier learning, this paper proposes a method of data dimensionality reduction to multi-scale Harr featrues with sparse representation theory, and chooses Naive Bayes classifier to learn and update the positive and negative samples of target. The sample with the maximum classification response value is selected as the current status of the target, that achieves fast and efficient tracking of moving target. Experimental results show that the method improves the real-time while maintaining a certain robustness and can apply to moving target tracking of visual robot.