【摘要】 针对存在更复杂运动模式的无序运动人群密集场景,提出了一种有监督的密集场景人群分群检测算法,不仅能够实现运动前景的自动定位和跟踪,而且还能够在有先验信息的条件下准确定位分群区域。结合个体之间运动轨迹时空距离特征、运动方向特征、格兰杰因果性特征、热能图特征以及运动相关性特征,采用结构化支持向量机在训练集上进行训练,对测试集人群进行了分组测试。选用了大量真实室内外密集人群无序运动场景视频进行测试,试验结果表明,该方法对密集人群分组有较好的处理效果。更多还原

【Abstract】 A supervised automatic group detection algorithm is proposed for complex movement pattern in crowds of disordered movement,which can locate and track movement automatically in foreground,and position the group area accurately with prior information. The spatiotemporal characteristics,movement direction characteristics,Granger causality characteristics,hot map characteristics and movement correlation characteristics of individual trajectories are combined,and the structured support vector machine( SVM) framework is used to train the characteristics to achieve the goal of crowd grouping. The algorithm combines various physical features and social characteristics,and makes the best use of the effective information in the group,making grouping more accurate.更多还原