【摘要】 针对车辆跟踪中背景复杂、快速运动以及干扰目标等问题,提出对车辆背景建模,利用目标信息与背景信息在目标跟踪中的不同作用达到在各种环境下鲁棒跟踪的效果。使用高速高精度的核化相关滤波算法,分别对车辆和车辆背景进行模型训练与在线更新。利用贝叶斯框架融合汽车特征预测结果和汽车背景特征预测结果。实验结果表明,基于背景建模的车辆跟踪方法能够提高跟踪精度,提高跟踪鲁棒性。更多还原

【Abstract】 Vehicle tracking is faced with complex background,fast motion,and disturbers. Background model is proposed for vehicle tracking. The target information and background information have different functions in object tracking. Exploring the relationship between the target and the background can promote robust tracking results. With kernelized correlation filters,the vehicle target and background models can be trained and updated efficiently. The vehicle target and background models are fused with a Bayesian framework. Experimental results show that the proposed vehicle tracking method based on background modeling can improve the tracking performance and robustness.更多还原