基于卷积神经网络与一致性预测器的稳健视觉跟踪

发布时间：2018-03-01 14:00

本文关键词： 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　出处：《光学学报》2017年08期 　论文类型：期刊论文

【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。
[Abstract]:Aiming at the problem of robust target tracking in video sequences, a visual tracking algorithm based on convolution neural network (CNN) and consistency predictor (CPN) is proposed, which constructs a dual input CNN model. The high-level features of frame sampling region and target template are extracted synchronously, the object and background regions are distinguished by logical regression method, the CNN is embedded into the CP framework, and the reliability of the classification results is evaluated by randomness test of the algorithm, and the reliability of the classification results is evaluated under the specified risk level. The classification results are outputted in the form of domain, the high credibility region is selected as candidate target region, and the global energy function in time and space domain is optimized to obtain the trajectory of the target. The experimental results show that the algorithm can adapt to the occlusion of the target. The appearance change and background interference are more robust and accurate than other tracking algorithms.
【作者单位】：西南科技大学计算机科学与技术学院;四川大学计算机学院;
【基金】：国家自然科学基金(91338107,91438119,91438120) 教育部博士点基金(20130181110095)
【分类号】：TP183;TP391.41

【相似文献】