基于深度学习的地铁客流量统计算法研究

Research on metro passenger flow detection and tracking algorithm based on deep learning

为提升地铁客流量统计的准确率及效率,针对地铁高密度人群多尺度、小目标检测需求,设计了一种改进的YOLOv3算法。为捕获深层抽象信息,增设卷积层加深网络结构,利用图像金字塔结构实现高低层特征信息融合,提升不同尺度目标检测精度;以IOU作为目标框及先验框误差的度量标准,实现高密度人群数据集的重聚,以提高小目标检测的准确率。实例分析结果表明,改进的YOLOv3算法与原YOLOv3算法相比,地铁客流量统计的准确率及效率均有明显提高。

In order to improve the accuracy and efficiency of subway passenger flow tracking and detection,according to the multi-scale and small target detection requirements of subway high-density people,this paper designs an improved YOLOv3 algorithm.In order to capture deep abstract information,a convolution layer is added to deepen the network structure,and the image pyramid structure is used to realize the fusion of high-level and low-level feature information,so as to improve the target detection accuracy of different scales.At the same time,IOU is used as the measurement standard of target frame and a priori frame error to realize the reunion of high-density population data set,so as to optimize the accuracy of small target detection.Finally,through experimental analysis,the improved algorithm has obvious advantages in accuracy and efficiency in subway passenger flow tracking and detection.