基于视觉图像踩踏风险预警研究

Risk Early Warning of Stampedes Based on Visual Image

在踩踏风险预警优化中,由于视频监控踩踏风险预警体系中,需要检测的区域较大,传统方法一旦区域过大,会出现大量的无关像素干扰,导致对监控图像数据的细节处理无法保证时效性,而且视频监控的数据量很多以及识别条件比较复杂,容易出现关键帧图像丢失、产生数据冗余和处理时间长等问题。提出采用粗糙集融合支持向量机的智能视觉技术的踩踏预警方法,通过小波变换及多尺度分析算法对采集视频数据的预处理进行修复缺失数据,利用粗糙集算法剔除多余属性,支持度理论进行算法约简,达到对视频数据处理的时间最短,最后用预警指标构建智能视觉系统预警模型,实现在大型群体活动中通过智能视觉技术对踩踏风险的预警。仿真结果表明,智能视觉技术应用于大型群体活动时,能够对踩踏事件进行有效预警。改进方法比传统方法能提高踩踏事件预警的时效性。

A risk early warning method of stampedes is presented based on the visual technology by using rough set fusion and support vector machine （SVM）. The video data are preprocessed to repair the missing data through the wavelet transform and multi-scale analysis algorithm, the redundant attributes are eliminated by using rough set algo- rithm, the algorithm is simplified by using the support theory, and the video data processing is achieved in the shor- test time. An early warning model of visual system is built with early warning indicators, and the risk early warning of stampedes in the large group activities is implemented by the vision technology. Simulation results verify the effective- ness of the proposed method.