复杂场景下的加权粒子滤波行人跟踪方法

A Weighted Particle Filter for Pedestrian Tracking in Complex Scenarios

针对粒子滤波跟踪算法在行人目标遮挡、光线干扰以及背景与行人相似等情形下,目标易发生漂移、跟踪精度不高的问题,本文提出一种加权粒子滤波行人跟踪方法。该方法联合遮挡模型和Online Boosting算法,利用在线学习实时更新强分类器,并结合跟踪时建立的遮挡模型,以及行人运动时与上一次目标位置的距离、相似度等影响因子,对粒子权重进行重新构造,实现了复杂变化场景下的行人自适应跟踪。对PETS-L2S1公共数据集和自有数据集的实验结果表明,本文提出的方法能有效去除目标遮挡、相似背景以及光线突变的干扰,实现稳定、准确、实时的行人跟踪。

Focusing on the problem that traditional particle filter tracking algorithm prone to drift and tracking accuracy is unsatisfactory when there is some interference from shading on the target, the light or which background is similar to the pe- destrian, a weighted particle filter for pedestrian tracking method is proposed. This method combines occlusion model and Online Boosting algorithm to reconstruct the particle weights, using online learning update strong classifiers in real time, meanwhile, combined with several impact factors, such as occlusion model, the distance and the similarity between last time target location and current location, to realize pedestrian adaptive tracking in the complex scenarios. Experiment re- sults on PETS-L2S1 public data and my own data set show that the proposed method can effectively remove the interference from object shelter, similar backgrounds and light mutation, the weighted particle filter method could accomplish pedestrian tracking stably, accurately and in real time.