融合HOG与ARMA模型的粒子滤波跟踪

传统的粒子滤波算法在跟踪目标受到相似背景干扰和遮挡或跟踪目标高速运动时，容易造成跟踪误差增大或跟踪失效的影响。针对室外运动目标跟踪的复杂性，提出一种对于干扰适应性较强的融合梯度方向直方图与自回归移动平均(ARMA)模型的粒子滤波跟踪方法。建立ARMA运动模型，用前两帧目标的位姿状态预测目标下一帧的状态，解决目标跟踪的角度变化与部分遮挡问题。实验结果表明，该模型能克服光照突变引发目标色彩突变的问题。

Particle Filter Tracking Combined with HOG and ARMA Model

The traditional particle filter tracking algorithm usually leads to tracking error or failure, when the target is interfered by the similar background or occlusion by the other object, and fast-moving environment. In order to track moving objects in complex outdoor scene, the particle filter tracking method with histograms of oriented gradients and Autoregressive Moving Average（ARMA） model fusion is proposed for its robustness. Various moving objects have different motion state in different scenarios, so a kind of ARMA motion model is introduced which forecasts the motion state of next frame based on motion states of former two flames, and overcomes the posture change and occlusion interference. Histograms of Oriented Gradients（HOG） observation model which can describe object＇s texture characteristic are applied to solve track problem in the complex scene that illumination is uneven or illumination suddenly change leads to mutation of object color.