作　　者： (高琳）; (王俊峰）; (范勇）; (陈念年）;

机构地区： 西南科技大学计算机科学与技术学院,四川绵阳621010

出　　处： 《光学学报》 2017年第8期222-231,共10页

摘　　要： 针对视频序列的稳健性目标跟踪问题,提出一种基于卷积神经网络(CNN)与一致性预测器(CP)的视觉跟踪算法。该算法通过构建一个双路输入CNN模型,同步提取帧采样区域和目标模板的高层特征,利用逻辑回归方法区分目标与背景区域;将CNN嵌入至CP框架,利用算法随机性检验评估分类结果的可靠性,在指定风险水平下,以域的形式输出分类结果;选择高可信度区域作为候选目标区域,优化时空域全局能量函数获得目标轨迹。实验结果表明,该算法能够适应目标遮挡、外观变化以及背景干扰等复杂情况,与当前多种跟踪算法相比具有更强的稳健性和准确性。 On the issues about the robustness in visual object tracking,a novel visual tracking algorithm based on convolutional neural network（CNN）and conformal predictor（CP）is proposed.A two-input CNN model is constructed to extract the high level features from the sampled image patches and target template simultaneously,and the logistic regression is used to separate the object from the background.The CNN classifier is embedded into the CP framework,and the reliability of classification is evaluated via algorithms randomness testing.The classification result with credibility is obtained by region prediction at a specified significance level.The image patches with high credibility are selected as candidate objects,thus,the target trajectory is obtained through spacetime optimization.Experimental results show that the proposed algorithm can adapt to the occlusion,target appearance changes and complex background,and it has a better robustness and higher precision than the current algorithms.

关 键 词： 机器视觉 目标跟踪 卷积神经网络 一致性预测器 时空域能量函数