Head Count for Business Intelligence Applications

Project Title

Head Count for Business Intelligence Applications

Project Area of Specialization Artificial IntelligenceProject Summary

The project comprises of several portions. Our main focus is to build a system that can count people in real time and give us the data so that we can use it whichever way we like in the best possible interest of the user.

We first take input from a live video camers i.e. CCTV camera and then we perform Head counting frame by frame. The data is timestamped and stored in our database which means that we know how many people were present at a certain place at a certain time. This data may be stored per second, minute, day, week, month etc.

Each frame is treated as a single image and using our trained CNN model we apply multiple head detection on that image. From here we get to know if there are heads present in that frame/image of the video. 

Using this data, we create a real-time statistical analysis based on current and previous data stored in our system. This data will be in front of the user in the form of Business Intelligence(BI).

Our BI will include real-time video stream, heat maps of people, statistical graphs and alert generation system that may warn the user that there may be more that usual of less than usual people present.

If the project is taken forward to a product form, based on the requirement of the user, we will provide multiple features in our BI which will be completely on the discretion of the consumer. 

Project Objectives

As stated earlier, the main goal is to count people in the best way possible hence we consider this our main and most important target to achieve. The algorithm we choose to count the number of people is of utmost importance and it will provide the best results.

The main objective of the project is to develop a system that is capable of counting the people from a video stream in real-time and then intelligently store the results in a database and generate alerts to help the user. The head counting is the main aspect which needs to be perfected to achieve good results in further parts of the project.

If we divide the objectives of the project into points we can get the following:

• Head counting model for real time video stream
• Business Intelligence

Head counting model will comprise the backend of our system. This will be responsible for providing authentic data frame by frame of the count of people present. Our database will also be present on the backend which will be timestamping and storing the data in real time.

The Business Intelligence is basically the front end of our project that will provide a Graphical User Interface (GUI) for the user. This will include several features that will help the user how to use the information provided to him regarding the strength of people in a certain area.

Some of the main features that we will provide are:

• Real time video stream of people being detected
• Heat maps of people
• Statistical graphs of current and previous head count data
• Alert generation based on abnormal head count

Project Implementation Method

We are using Machine Learning models to count the head of person standing in the queue on any public place like hospital, restaurants, airports and bank etc.  We started our project with the literature review of different head detection models and after the literature review, we moved on to the gathering of dataset and after modification of dataset we moved on to the testing of models. From the literature review we selected a model known as Eldar Pose Estimation model and we moved on to fulfill the prerequisites for the selected model and stated to test the model on already available and our prepared dataset.

Selecting this model was not quite successful because this model was not giving the accurate results of the Head count on both type of datasets. After that we rejected this model and we moved on to another model know as Faster RCNN. This model was more accurate than the Eldar Pose Model but it also has some limitations. This model was using the technique of “Regional-Convolutional Neural Network (RCNN)” which basically is a state-of-the-art visual object detection system that combines bottom-up region proposals with rich features computed by a convolutional neural network. We modified this model according to our requirements and then we used this model for the still images first and after getting required results we moved onto the video testing of this model.

Now currently we are using Mobile-Net CNN for the head counting purposes. Till now we have developed heatmaps on the video stream to check the density of humans on the current frame and also developed time-stamped statistical graphs for the counted number of people.

We have distributed our project into different major and minor tasks. The list of major tasks is given below:

1. Literature review
2. Dataset Gathering
3. Dataset Preparation and Exploration
4. Model Selection and Modification
5. Training of Model
6. Heatmap Generations
7. Database Creation
8. Stats Generation
9. Development of GUI
10. Project Report

Benefits of the Project

We are looking to provide the business community with a reliable solution to the everlasting problem of queue mismanagement and mishaps in public administration that occurs on regular basis in their organizations, this research and software/hardware based project aims to be a pivotal game changer in this regard. This project hopes to utilize the latest techniques of Deep Learning a subset of Machine Learning hence minimizing the human involvement in public administration. By counting the number of people present in a certain area where queues and counters are functioning, time stamping that number, and storing it in the database, we’ll be able to use that collected data in order to generate statistics, and after studying the trends for a certain period, we can make predictions for upcoming near future, also generating alerts for anomalies. All these business intelligence terminologies will be embedded in a Graphical User Interface that is the front end of our final product while back end being the Deep Learning trained model and Database. A onetime investment for the organizations which shall reimburse its cost within a very quick time frame in the form of lower expenditure on human resources and better profits generation from satisfied customers.

Technical Details of Final Deliverable

The final deliverable of the project will be a web application based system which will be running on a Raspberry Pi 3b. This will be our product form which will have two portions:

1. Front-end
2. Back-end

The front-end will be made as a web application using NodeJs or any other form of front-end development tool. It will comprise of a display which will have the following features:

• Display of live video stream with people count in real-time (Deep learning)
• Display heatmaps of the same video stream (Computer vision)
• Display statistical graphs to show headcount over the day/week/month
• Display of some statistical results in the form of future predictions for user to read and make changes to staffing per need
• Alerts generation to help improve staffing

The back-end of our project will include the following:

• Headcount algorithm which will count number of people in a frame in real-time
• A database, most prefferably made on MySQL, to keep all the time-stamped data of number of people for future use and predictions

Final Deliverable of the Project HW/SW integrated systemCore Industry OthersOther Industries Education , IT , Medical , Food , Manufacturing Core Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development Goals Decent Work and Economic Growth, Industry, Innovation and InfrastructureRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	65000
Custom PC for training of algorithms	Equipment	1	60000	60000
Raspberry pi 3b	Miscellaneous	1	5000	5000