基于信道状态信息的室内人群计数方法

Indoor Crowd Counting Based on Channel State Inform

人群计数对于室内空间公共安全管理、建筑节能优化等都具有重大的价值,商场也可以根据人数信息进行商品推荐和流量调控。传统基于视觉图像的方法部署成本高,受视线遮挡严重,而且容易造成隐私问题。采用商用WiFi的信道状态信息(CSI)来进行室内人群计数。首先对原始信号数据进行预处理,最小化噪声并降低数据复杂度;然后,通过滑动窗口将无线时间序列信号转换为热图图像,并设计了一个卷积神经网络CNN对热图进行特征提取,用以映射相应环境下的不同人数。实验设计了一种人员位置相对静态的模拟办公场景和另一种人员走动的模拟商业场景进行验证,结果表明所提方法在静态和动态条件下的准确率分别达到了98%和89%,相比传统算法均取得了更优效果,证明了该方法的有效性。

Indoor crowd counting is of great value for public safety protection, smart building system optimization, etc, and in many shopping malls, product recommendations and flow regulation can be carried out based on the number of people. However, traditional visual image-based methods are expensive to deploy, severely obscured by the line of sight, and easy to cause privacy issues. Commercial WiFi channel state information(CSI)is uesed to count indoor people. According to the fact that neural network can better process the image features, firstly, the original signal data is preprocessed to minimize the impact of noise while reducing data complexity, then, in order to highlight the signal characteristics, a sliding window is used to convert the preprocessed signal into a heat map in the time dimension. Besides, a convolutional neural network(CNN)is designed to extract the features of heat maps to map different numbers of people in the corresponding environment. In the experiment, a simulated office scenes with relatively static personnel positions and another business scenes with people walking around are designed for verification. Experimental results show that the accuracy of the proposed method reaches 98% and 89% at static and dynamic scenes, respectively. It has achieves better results compared with the traditional algorithm, proving the effectiveness of the proposed method.