kandi background

Price recommendation with people counting kit

by suriya

Restaurant Business Analytics for the price, discount, etc prediction with people counting web app Innovative solution for gaining profit in standalone startup restaurants. According to many reports, the main factor that determines a restaurant's profit is the people count. So we have proposed a solution where a restaurant's past history of prices, dishes, and such factor's transaction has been taken into account and providing a recommendation for the price to be fixed for a particular day, the dish quantity to be prepared, etc. Trend analysis is much more important, for that we use tableau, which provides real-time public inference on data, here the owners can check their past transaction and analyze the trend. Use the open source, cloud APIs, or public libraries listed below in your application development based on your technology preferences, such as primary language. The below list also provides a view of the components' rating on different dimensions such as community support availability, security vulnerability, and overall quality, helping you make an informed choice for implementation and maintenance of your application. Please review the components carefully, having a no license alert or proprietary license, and use them appropriately in your applications. Please check the component page for the exact license of the component. You can also get information on the component's features, installation steps, top code snippets, and top community discussions on the component details page. The links to package managers are listed for download, where packages are readily available. Otherwise, build from the respective repositories for use in your application. You can also use the source code from the repositories in your applications based on the respective license types.

Data gathering and cleaning

Handpicked data of restaurant prices, dish count, etc, and preprocessing using pandas and NumPy would give us clean data for modelling, the data is raw, formatting for our use is the key to good accuracy.

People Counting

After done with the data, we can use the pre-trained model and tune it to do non-max suppression to get prediction only for people. And get predictions as the people count, in selected areas. Tracking is necessary to know when people come into the frame and track them till they move out, for that Dlib and OpenCV libraries are used.

Recommendation and Time series

The main idea here is to recommend prices to restaurant owners with respect to the people count for the future. So with the cleaned data, we model it using the prophet model from the Facebook research do time series analysis i.e to predict the future values and then apply dlrm a recommender system model and recommend people the right dish with respect to the past history, and recommend a price to the owners with respect to the people count.

Deployment

The detection model(counting model) with the price recommendation model, all wrapped and connect as a single application(a web app). Streamlit is an easy-to-use deployment framework, particularly for machine learning.