Thanks to the development of the Internet of things (IoT) and edge computing, the smart cameras across cities provide a massive amount of image samples with time and location labels, laying a solid basis for deep mining of image information and in-depth decision analysis. Therefore, this paper proposes to convert the images with spatiotemporal labels into quantifiable data on emotions, and apply them to crime prediction. Firstly, human emotions were divided into three categories: negative, neutral, and positive. Then, facial expression recognition (FER) was employed to quantify the portrait data. The emotion features thus acquired were imported to the crime prediction model, enhancing the model’s explanatory power. Finally, our method was compared with kernel density estimation (KDE) on six typical crimes. The results show that introducing emotion data helps to reveal the interaction between emotions and crimes and improve the performance of crime prediction.

得益于物联网（IoT）和边缘计算的发展，城市中的智能相机可提供大量带有时间和位置标签的图像样本，为深度挖掘图像信息和进行深入的决策分析奠定了坚实的基础。 。因此，本文提出将时空标签图像转换为可量化的情感数据，并将其应用于犯罪预测。首先，人类情感被分为三类：消极，中立和积极。然后，使用面部表情识别（FER）来量化肖像数据。这样获得的情感特征被导入到犯罪预测模型中，从而增强了模型的解释能力。最后，将我们的方法与六种典型犯罪的核密度估计（KDE）进行了比较。