基于粒子滤波的多特征融合视频行人跟踪算法

针对车载视频行人跟踪问题，提出一种基于粒子滤波框架下的多特征融合跟踪算法。为克服车载视频中行人运动与摄像机运动产生的非线性和非高斯性，采用基于蒙特卡罗抽样的粒子滤波跟踪算法，使用一阶自回归动态模型预测目标状态，观测模型自适应加权融合的4种互补性特征。实验结果表明，与没有粒子滤波和多特征融合的跟踪算法相比，在相同精确率水平上，该算法的召回率提高20%以上。

Pedestrian Tracking Algorithm in Video of Multi-feature Fusion Based on Particle Filter

This paper presents a tracking algorithm based on multi-feature fusion in the particle filter framework to solve the problem of pedestrian tracking in onboard videos. To deal with the nonlinearity and non-Gaussianity caused by the motions of the pedestrians and the cameras in onboard videos, the particle filter tracking algorithm based on Monte-Carlo sampling is employed, the targets’ states are predicted by first-order self-regression dynamic models, and the observation model is proposed to fuse four complementary features. Experimental results show that the recall of the proposed algorithm improves by more than 20% at the same precision level than the tracking algorithm without particle filter and multi-feature fusion.