基于粒子滤波与样本加权的压缩跟踪算法

该文针对压缩跟踪算法无法适应目标尺度的变化以及没有考虑样本权重的问题，提出一种基于粒子滤波与样本加权的压缩跟踪算法。首先，对压缩特征进行改进，提取归一化矩形特征用于构建目标表观模型。然后，引入样本加权的思想，根据正样本与目标之间距离的不同赋予正样本不同的权重，提高分类器的分类精度。最后，在粒子滤波的框架下融合尺度不变压缩特征进行动态状态估计，在粒子预测阶段利用2阶自回归模型对粒子状态进行估计与预测，借助观测模型对粒子状态进行更新，并且对粒子进行重采样以防止粒子退化。实验结果表明，相比于原始压缩跟踪算法，改进算法能够更好地跟踪目标尺度的变化，提高跟踪的稳定性和准确性。

Compressive Tracking Algorithm Based on Particle Filter and Sample Weighting

To solve the problem that Compressive Tracking (CT) algorithm is unable to adapt to the scale change of the object and ignores the sample weight, an optimized compressive tracking algorithm based on particle filter and sample weighting is presented. Firstly, the compressive feature is improved for building a target apparent model with normalized rectangle features. Then, the thought of sample weighting is utilized. In order to increase the precision of the classifier, different weights are given to the positive samples in accordance with the different distances between the positive samples and the object. Finally, the dynamic state estimation is made under the particle filter frame with integrating the scale invariant feature. At the phase of particle prediction, a second-order autoregressive model is utilized to obtain the estimation and prediction of the particle state. The particle state is updated with the observation model. The particles resampling is used to prevent the degradation of particles. Experimental results demonstrate that the improved algorithm can adapt to the scale change of object, and the accuracy and stability of the compressive tracking algorithm is improved.