基于时间序列模型的粒子滤波行人跟踪算法研究

Particle Filter Tracking Moving Pedestrian Bsed on Adaptive Time Series Models

摘要:

针对行人运动的随机性导致运动状态模型适应性差和人在行走过程中可能发生短时全部或局部遮挡导致行人跟踪算法精度较低的问题,提出基于时间序列模型的粒子滤波行人跟踪算法；建立了行人运动时间序列模型；给出了基于对视频序列初始帧的检测,确定行人的位置、宽高等作为跟踪先验信息的方法；由先验信息计算加权颜色直方图构建初始粒子群分布,并利用时间序列运动模型预测粒子在下一时刻的状态分布,并更新粒子权值；根据有效粒子的个数判断是否进行重采样；最后由所有粒子的加权和估计行人的运动状态；仿真实验表明:文中提出根据行人的运动轨迹时间序列运动模型可使行人的状态估计更准确,预测误差进一步减小,预测精度得到了提高。

Abstract:

Aiming at the poor adaptability of the movement state model because of the random pedestrian movements and the less pedestrian tracking algorithm accuracy caused by the short-time wholly or partially obscured movements of the pedestrian. First, the time series model of pedestrian movement is established, then the method of the achievement of the tracking the prior information is established by detecting the initial frame of the video sequence to determine the location and the wider, and etc. The weighted color histogram is calculated based on the prior information to the initial particle size distribution, and the distribution of the next time particle state is predicted by use of the time-series model, the particle weights is also updated. Whether resembling or not is depend on the number of effective particles. Finally, the pedestrian motion state can be estimated based on the weighted sum of all particles. Experiments show that improved particle filter algorithm allows a more accurate estimate of the pedestrian movement .