摘要： 针对基于单目视觉的公交乘客人数统计判定方法不稳定、计数结果不准确的现状,结合公交车门附近乘客运动行为的复杂性和多样性,以及乘客运动行为对计数判定方法的干扰,给出一种基于乘客多运动行为分析的计数判定方法。采用轨迹聚类的方式对乘客运动行为进行分析,结合轨迹的空间特征和方向特征计算轨迹距离,并使用层次聚类方法进行聚类。分析聚类结果中每一类别所对应乘客类的运动行为,讨论各乘客类的运动行为对常用计数判定准则的影响,由此提出一种改进的公交车客流计数判定方法。利用采集的乘客上下公交车视频图像进行实验,结果表明,该方法能获得较高的统计精度和较好的稳定性。

关键词: 乘客人数统计, 轨迹聚类, Hausdorff 距离, 层次聚类, 乘客运动行为, 计数判定

Abstract: In view of the instability and inaccuracy of bus passenger statistic with monocular vision based method,a new statistic method based on passengers’ multi-movement behavior is proposed,which combines the complexity and variety of passengers’ behavior. Passengers’ movement behavior is analyzed with trajectory clustering algorithm. The trajectory distance which is clustered with hierarchical clustering method is calculated according to the spatial feature and the directional feature of the trajectory. The clustering result corresponds to a certain kind of movement behavior of passenger,the influence of each movement behavior to the common counting criterion is discussed. In a result,the method based on passengers’ multi-movement behavior for bus passenger statistic is proposed. Experimental results based on the video when passengers getting on and getting off show that the method can obtain a high statistical precision and good stability.

Key words: passenger number statistics, trajectory clustering, Hausdorff distance, hierarchical clustering, passenger movement behavior, counting criteria

中图分类号: 

• TP391