Research On Intelligent Dining System Based On Image Recognition

Smart Dining refers to a smart catering service based on Internet of Things,cloud computing and artificial intelligence technology,which can significantly reduce the labor force and operating costs with new technologies and algorithms.Smart catering services have been used in various fields of catering scenario with the rapid development of China's catering industry.However,the previous ones are basically used in social catering,and only a few smart catering service systems are suitable for large-scale group dining halls.The canteen systems are not specially designed for large-scale scenario usually lead to congestion in rush hours due to relatively concentrated dinning time and the low level intelligence.In view of the above shortcomings,we proposed a set of Image Recognition Algorithm based smart food catering system.The main work of this paper is as follows:(1)Aiming at the characteristics of group dining hall,such as the dense number of people and the concentrated dining time,a fast image recognition based settlement system is built.According to the sale mode of "one dish for one price" adopted by the self-selected canteen,the information such as tableware and price are imported into the system in advance,and the image recognition technology is used to verify the recognized tableware type with the ones stored in system,and then the total price of the meal is calculated to complete the settlement.(2)A threshold segmentation method based on HSV is used to segment the tableware.We first use a channel wise threshold segmentation in HSV color space to segment the tableware,then a region-based denoting algorithm is applied to fill the hole in the segmentation mask.In order to improve the tableware segmentation performance,we further introduce an algorithm named MBS to analyze the significance spectrum of the original image,and use the intersection ratio relationship between the significance spectrum and HSV based segmentation mask to extract the final tableware segmentation mask.The the proposed method can not only avoid over-dependence on tableware significant spectrum information,but also make up for the deficiency of robustness in threshold segmentation algorithm.(3)An Image Recognition Algorithm based SSD with binary images of tableware information as input are proposed.The proposed method can effectively shield the influence of different dishes and background information on the SSD model,which shows significant performance gain compared with the model trained on original image.(4)To resolve the low degree of informatization issue in large group dining halls,we design a Django framework based data management website.The website mainly includes four modules: account management,user management,user information query and system management.The management website can collect the information of consumers and present the data with the format of summarized chart to the canteen administrators.In this paper,we proposed a set of Image Recognition Algorithms based smart food catering system for large-scale group dining halls.The proposed system can effectively alleviate the congestion pressure during peak time and collect catering data.The proposed system has shown its theoretical and practical significance during pilot run.