Design And Implementation Of Visual Perception System Based On Microservice Applied To Wisdom Store

With the advent of the Internet information age,wisdom stores,which focus on meeting consumer demand and improving efficiency,are rising gradually.In offline platform,how to make full use of video image information to analyze customers'interest,explore potential value and guide commodity planning has become an increasingly concerned issue for merchants.At present,computer vision technique is developing rapidly.The visual perception system,which takes object detection technique and attribute recognition technique as the core,can detect customers and recognize basic attribute characteristics from surveillance video information,providing help for the merchants to develop service strategies.However,there are still some problems in visual perception system that need to be solved urgently:(1)the quality of surveillance video in complex scenes such as shopping malls and supermarkets will be affected by lighting condition,which will interfere with detection and recognition;and(2)object detection algorithm should ensure real-time performance as well as ensure accuracy;and(3)the result of attribute recognition algorithm should be as accurate as possible;and(4)the performance of object detection algorithm and attribute recognition algorithm based on deep learning method will be different in different scenarios.In view of the above problems,the main work and research of this paper are as follows.(1)Expounding the research significance,this paper introduce the research status of object detection technique and attribute recognition technique and introduce relevant technique principles of the visual perception system.(2)Compared with the existing image enhancement technique,in consideration of the scenes requirement,this paper introduce an image enhancement method based on parabolic function,which ensures real-time and performance well in image processing.(3)Compared with the existing object detection algorithm,in consideration of the scenes requirement and the YOLOv3 algorithm has poor performance in the detection of small targets,this paper introduce an improved algorithm based on YOLOv3 algorithm,which uses the dataset of the supermarket scene for training and testing.The improved algorithm base on YOLOv3 algorithm has better performance in shopping malls and supermarkets scene,whose precision improved by 2%and recall impro ved by 3%.(4)Compared with the existing attribute recognition algorithm,in consideration of the scenes requirement,this paper introduce an attribute recognition algorithm based on attention mechanism,which uses RAP dataset for training and testing.The attribute recognition algorithm based on attention mechanism has a good performance in the scenes of shopping malls and supermarkets.(5)The visual perception system applied to wisdom store is designed and implemented,and based on the analysis of functional requirements and performance requirements,this paper design the system architecture based on microservice.According to functions,the system is divided into five microservice modules:the user management module,the image capture module,the image preprocessing module,the object detection module and the attribute recognition module.The system is implemented by flask framework,and the algorithm designed in this paper is applied to implement three core functions of image enhancement,object detection and attribute recognition.Each of the system modules is independent and the whole system is extensible.The visual perception system provides a convenient platform for merchants to improve service.