基于图像处理的公交车内人群异常情况检测

摘要

To effectively monitor abnormal crowd behavior in the specific environment of bus,a method of detecting abnormal condition on bus based on improved image processing algorithm was proposed.Regions of interest for video images were got and those images were pre-processed.Moving targets were extracted using an improved ViBe algorithm and recognition regions were located through multi-scale moving windows.Several recognition regions of continuous frames were combined to detect abnormal crowd behavior based on improved convolutional neural networks algorithm.Whether the crowd behavior was abnormal on bus was determined through the recognition results.Compared with traditional method and traditional convolutional neural networks algorithm,it can be known that the proposed algorithm has higher detection accuracy of 93.5％ and lower error rate of 1.6％,which has a high reference value in the practical application.%为有效地监控公交车这一特定环境中人群的异常行为,提出一种公交车内人群异常情况检测的方法.对视频图像确立感兴趣区域,进行预处理;通过改进ViBe算法提取运动目标,引入多尺度滑窗算法确定识别区域;结合连续多帧识别区域进行改进卷积神经网络算法的异常行为识别,通过识别结果判断公交车内人群是否异常.与传统方法的比较结果表明,该算法的检测正确率较高,可达93.5％,误检率较低,仅为1.6％,在实际应用中具有较高的参考价值.