元胞驻留视频群体状态预报方法

Inner-ellular automata method for video pedestrian state forecasting

视频场景人群状态预报是防止公共场所安全事故的一种重要措施.为此提出了一种视频场景群体状态预报模型.该模型使用格内驻留元胞自动机(inner-grid parking cellular automata,IPCA)模型预报视频中群体状态.首先,通过光流法进行运动跟踪作为IPCA的输入特征对场景进行建模,并根据特征点的运动状态自适应地调整其在元胞格内的驻留时间以提高预报精度;其次,通过分析IPCA模型的预测结果判断群体微团间的相互作用得到视频帧的状态信息,从而达到对场景中人群状态进行预报的目的;最后,利用当前场景的状态信息对该场景的预报输出进行反馈校正,使模型能够及时地反映场景中群体状态的变化,该模型不依赖于具体的视频帧,以微团结构的碰撞作为异常检测依据.实验结果表明,与传统的检测方法相比,该预报模型能够提前预报视频场景中的异常状态,并具有较好的准确性.

Forecasting pedestrian states with the help of video analysis is a significant mean to prevent accidents in public places.An improved Inner-grid Parking Cellular Automata（ IPCA） model is proposed to forecast pedestrian state.Motion tracking and scenario modeling are firstly achieved by optical flow method,and then motion states of pedestrian are used to adjust parking time in cellular adaptively to improve forecasting accuracy.Here,the state of an incoming video frame is acquired by analyzing the interaction between micelles,which is provided by the forecasting result of IPCA.After that,a feedback algorithm is employed to revise the forecasting result so that the model could reflect precisely the change of pedestrian state in the scenario.A criterion is also proposed to judge the abnormal state of a video frame,which is collision between two micelles.Compared to other traditional detection methods,IPCA based model has a good ability in predicting the abnormal state ahead of time and obtaining better accuracy.