【摘要】 为适应长期跟踪的需求,提出一种基于特征补偿的单目标跟踪方法。通过分析现有算法的优缺点,使用低层特征的相关滤波模型完成简单场景的任务,以保证速度;使用高层语义特征的卷积神经网络模型完成复杂场景的任务,以提升精度和鲁棒性;引入一个简单的分类器,作为切换特征的标志,有选择的对模型的模版进行更新,降低累计误差。结合实例验证了算法的有效性,在保证较快速度的同时,精度和鲁棒性均有较高的提升。更多还原

【Abstract】 To meet the requirements of long-term tracking,a single target tracking method based on feature compensation was proposed.The advantages and disadvantages of the existing algorithms were analyzed,the low-level feature correlation filtering model was used to complete the tasks of simple scenes to ensure the speed.The convolutional neural network model with highlevel semantic features was used to complete the tasks of complex scenes to improve the accuracy and robustness.A simple classifier was introduced to mark the switching feature,and the model template was updated selectively to reduce the cumulative error.An example was given to verify the effectiveness of the proposed algorithm.It ensures high speed and high accuracy and robustness.更多还原