基于改进的粒子滤波的红外视频行人跟踪

Pedestrian Tracking in Infrared Video Based on Improved Particle Filter

摘要:

提出了一种基于改进的粒子滤波的红外视频行人跟踪算法，实现了在传统粒子滤波算法的框架下，使用有向梯度直方图(histograms of oriented gradients，HOG)来描述跟踪目标的特征.算法在粒子权值和相似度计算中使用HOG，替代现有的颜色空间欧式距离测度，克服了红外视频中颜色信息缺失的困难.试验表明，与传统的粒子滤波算法相比，本文算法更能准确有效地跟踪复杂场景中的行人，提高了跟踪的鲁棒性.

Abstract:

An improved particle filter method for pedestrian tracking in infrared video is proposed. The objects are described in the scheme of particle filter using Histogram of Oriented Gradients(HOG). Instead of the Euclidean distance in color space, the HOG is employed to describe the similarity and compute the weights of the samples, which solves the issue of lack of color information for infrared video. Experimental results show that the method is more accurate and effective tracking of moving targets in complex scenes than traditional particle filter algorithm in infrared video.