【摘要】 提出了一种基于模型融合的行人跟踪算法。该算法在粒子滤波框架中融合了基于颜色直方图特征的产生式模型以及基于HOG特征的鉴别式模型,并结合离线学习和在线互学习对HOG鉴别式模型进行更新。实验结果表明,该方法能够有效地适应跟踪过程中的光照变化、姿势变化以及部分遮挡。更多还原

【Abstract】 A human tracking algorithm is proposed in this paper,which combines one generative color histogram model and one discriminative HOG model.The discriminative HOG model is further updated by offline learning and online co-training.Experimental results show that this algorithm could effectively handle various conditions such as illumination change,pose change and occlusion with great robustness.更多还原