基于卷积神经网络的监控视频人数统计算法

A surveillance video crowd counting algorithm based on convolutional neural network

监控视频中人数统计是现代安防的重要任务之一,具有较高的研究意义和应用价值.虽然近年来取得较大的进展,但仍无法很好地解决监控场景人数统计精度、高清图像耗时问题.为此,作者提出一种基于卷积神经网络与岭回归联合的人数统计方法.通过卷积神经网络回归图像中人头中心点获得人群密度分布特征图,然后使用岭回归模型分析人群密度分布特征图得到该帧图像对应的人数.作者提出的算法通过在多组视频图像上进行了测试,并与经典算法做了比较.实验结果验证了作者方法的有效性.

Crowd counting in surveillance video is one of the important modern security tasks with high research significance and application value. It has made great progress in recent years, but still has not solved the problems about accuracy of crowd counting in surveillance video and time consuming of high resolution images. Therefore, the crowd counting algo- rithm by incorporating convolutional neural network and ridge regression was proposed in this paper. We could get the crowd density map from regression of the centers of heads through convolutional neural network, then used ridge regression model to analyze the crowd density map to got the number of people in current frame. The proposed algorithm had been tested on several videos and compared with several classical algorithms. The experimental performance validated the effectiveness of the proposed method.