基于颜色粒子滤波的多目标跟踪器设计

Design of Multi-object Tracker Based on Color Particle Filtering

针对多目标跟踪中的目标遮挡和身份切换问题,在颜色粒子滤波的基础上,将单目标跟踪器扩展为多目标跟踪器。当目标轨迹彼此较接近时,利用自适应冲突预防模型分离临近轨迹。当目标发生遮挡时,跟踪器的整体模型会被分割成多个部分,使用可见部分执行跟踪和遮挡推理。当目标外观相似时,使用轨迹监测方法处理遮挡情况。当目标发生完全遮挡时,重新初始化遮挡者周围的粒子并采集目标,从而实现多目标跟踪。实验结果表明,与MIT和LSAM跟踪器相比,该跟踪器在假阴性率、假阳性率、误匹配率和多目标跟踪准确率方面均具有明显的性能优势。

Aiming at the problems of occlusion and identification switching in multi-object trackeing, an extension scheme from a single-object tracker to a multi-object tracker is proposed based on color particle filtering. When the trajectories are very close to each other, the proposed adaptive conflict prevention model is utilized to separate trajectories close to each other. When object occlusion occurs, the overall model of the tracker is divided into several parts, and the visible parts are used to execute tracking and occlusion reasoning. When the objects are similar in appearance, monitoring method is applied to handle the occlusion. When the object is completely occluded, the particles around the occluded objects are re-initialized to collect the re-emergencing object and realize multi-object tracking. Experimental results show that the proposed tracker has obvious advantages in the aspect of False Negative Rate （FNR）, False Positive Rate （FPR）, Mis-matching Rate（MMR）, and Muti-object Tracking Accuracy （MOTA） compared with MIT and LSAM trackers.