基于微服务的智慧门店视觉感知系统设计与实现

Design and Implementation of Visual Perception System Based on Microservice Applied to Wisdom Store

根据智慧门店场景应用需求,设计并实现基于微服务的视觉感知系统。按照功能需求将系统划分为用户管理、图像采集、图像预处理、目标检测4个微服务模块,微服务模块以接口形式提供服务,具有独立性。使用暗光增强方法对低照度原始图像进行图像增强,改进YOLOv3网络并在自制的商场超市场景下的人群数据集上进行训练,从而提高视觉感知系统在智慧门店场景下的检测表现。实验结果表明,通过在自制商场超市场景下的人群数据集上进行测试,改进的检测网络较原YOLOv3网络的检测精确率提高了2%,召回率提高了3%。

According to the application requirements of wisdom store,the visual perception system based on microservice is designed and implemented.According to the functional requirements,the system is divided into four micro-service modules:user management,image capture,image preprocessing and object detection.The microservice modules which are independent provide service in the form of interface.The image enhancement method is used to enhance the original image of low illumination.The YOLOv3 network is improved,which uses the self-made dataset of the supermarket scene for training,so as to improve the detection performance of the visual perception system in the wisdom store.The experiment result shows that the detection precision is improved by 2%and recall improved by 3%.