【摘要】 马尔科夫决策多目标跟踪算法在目标遮挡和目标消失再出现等情况下容易丢失目标,无法进行持续有效的跟踪。利用核相关滤波器较强的判别能力,提出了一种基于核相关滤波优化的马尔科夫决策在线多目标跟踪算法。结合核相关滤波器增强跟踪目标的特征表达,先对跟踪点集合进行改进,在均匀分布点的基础上加入目标表达能力更强的Harris角点;设计一个归一化交叉相关相似性分数,利用在线学习存储的目标模板对中值流(Median Flow)跟踪和核相关滤波(Kernel Correlation Filter,KCF)算法预测的目标位置进行分析,优化最终目标输出位置;通过在核相关滤波器加入高置信度模型更新策略,结合中值流跟踪稳定性作为跟踪器评判标准,判断目标继续保持跟踪还是转为丢失状态。在2D MOT 2015测试集与马尔科夫决策(Markov Decision Process,MDP)跟踪器进行比较,主要跟踪性能参数有显著提高。更多还原

【Abstract】 The Markov decision multi-target tracking algorithm is easy to lose the target in the case of target occlusion and reappearing of the target, and it is impossible to perform continuous and effective tracking. Based on the strong discriminative ability of the kernel correlation filter, this paper proposes an online multi-target tracking algorithm based on kernel correlation filter optimization. Combining the kernel correlation filter to enhance the feature representation of the tracking target, the authors first improved the tracking point set, and added the Harris corner point with stronger target expression ability based on the uniform distribution point. Then a normalized cross-correlation similarity score was designed, the target template stored in the online learning store was utilized, and the target position predicted by the Median Flow tracking and Kernel Correlation Filter(KCF) algorithm was analyzed to optimize the final target output position. By adding the model update strategy with high confidence in the kernel correlation filter, and taking the median flow tracking stability as a tracker evaluation criterion, we determined whether the target continued to track or lose to a lost state. In the 2 D MOT 2015 test set compared with the Markov Decision Process(MDP) tracker, the main tracking performance parameters were significantly improved.更多还原