基于惯常速率的多类目标轨迹预测

Trajectory prediction for multi-class target based on preferred speed

为了解决监控视频中对大量不同类型的运动目标进行运动轨迹预测的问题,系统地提出了对多类目标的轨迹预测流程;在社交力模型的基础上,探讨了一种新的以惯常速率为聚类对象的运动模式特质分类方法,并应用这种方法在Stanford Drone数据库上取得了领先的结果.该方法可以使目标轨迹预测的研究对象拓展到除行人以外的其他任何移动目标,如汽车、自行车等运动物体,并对它们的运动轨迹进行有效预测.该方法在实现高精度预测的基础上,极大地缩短了目标分类所用的时间,分类效率的提高达5个数量级.

To solve the problem of trajectory prediction for multi-class target,Firstly,a systematic procedure to predict the trajectory for multi-class target was proposed.Secondly,a brand-new target classification method based on social force model was introduced,which clustered the targets with preferred speed.The proposed algorithm obtained state-of-art results on Stanford Drone Dataset.The classification method could be applied to all kinds of moving targets in videos,including cars and cyclists,for trajectory prediction.The efficiency of target classification was improved by five orders of magnitudes.