Smart social distancing detector for higher educational institutions

The coronavirus or COVID-19 is a deadly virus which spreads easily through physical contact from person to person. The widely recommended ways to prevent it include washing of hands, using face masks and maintaining physical distance also known as Social distancing. Social distancing significantly r

Project Title

Smart social distancing detector for higher educational institutions

Project Area of Specialization

Artificial Intelligence

Project Summary

The coronavirus or COVID-19 is a deadly virus which spreads easily through physical contact from person to person. The widely recommended ways to prevent it include washing of hands, using face masks and maintaining physical distance also known as Social distancing. Social distancing significantly reduces the risk of infection by avoiding crowded places and many countries have introduced social distancing as a basic precautionary measure. However, its implementation and monitoring is a hard and challenging task as it is extremely difficult to monitor each and every person in various places like malls, restaurants, airports, train stations, educational institutes etc. Educational institutes around the world are looking for smart solutions for monitoring their faculty and students. However, it is not easy as educational institutes have a large faculty and student body which move around the various places of a typical educational institute. These places include classrooms, auditorium, exam halls, front desk and other allied departments. In order to ease this difficulty, Artificial Intelligence AI based software solutions have been proposed in recent times. These systems use the image processing and machine learning techniques to identify and monitor people who are not following the social distancing guidelines of maintaining a distance of 6 feet apart. The aim of this project is to compare and evaluate the existing solutions in the literature to overcome the existing limitations of the recently developed systems. Moreover, the objective is to develop an AI based smart social distance monitoring system which fulfills the basic needs of a higher educational institute with advanced monitoring and near to real time analytics features. The proposed project will be able to help higher educational institutes to implement and monitor social distancing in a smart way using advanced AI and computer vision techniques.

Project Objectives

• This project aims at monitoring people violating Social Distancing over live video footage coming from CCTV Cameras. Uses YOLOv3 along with DBSCAN clustering for recognizing potential intruders. A Face Mask Classifier model (ResNet50) is trained and deployed for identifying people not wearing a face mask. For aiding the training process, augmented masked faces are generated (using facial landmarks) and blurring effects (frequently found in video frames) are also imitated.
• The proposed project will only detect moving objects i.e walking humans
• It will only detect social distancing in predefined areas of HEIs
  1. The  objective is to develop an interactive dashboards which enables advanced   analytics of social distancing
  2. The proposed system will be able to perform crowd analysis at various areas(zones) of the Institute
  3. The system will allow Tracking of objects(student, staff, faculty) at different zones
  4. It will enable Colour coding for high, medium and low risk students
  5. Timely alerts feature will be provided
  6. Hourly, daily and weekly reports of the violation will be generated by our proposed system
  7. Contact tracing feature using the multiple cameras at different zones will be provided. 

Project Implementation Method

 

Our proposed method will be trained using Keras' Image Data Generator with augmentation. Training data now keep changing after each epoch. Rotation, Horizontal Flip, Brightness Shift are some of augmentations that can be applied. We also incorporate our blurring augmentation with some associated probabilities during the training. The model needs to be trained on tons of relevant data before we can apply it in real-time and expect it to work. It needs a lot of computational power and I mean a lot! We can try our models trained on a small dataset in our local machines, but it would not produce desirable results. We can use pretrained open-source models for now. We will use pretrained ResNet50 Model with a modified top. Some optimizations can be made in the form of vectorization. For getting the position of a person, there are various approaches. One of them being simply using the centers of the bounding boxes, the one used in this project. Other one is using OpenCV's perspective transform to get a bird's eye view of the positions, but that kind of needs pretty accurate frame of reference points. Using it also increases the complexity of the system by a bit. However if implemented correctly, it will no doubt produce better results. 

Benefits of the Project

This is an advance version of vision based social distancing detection system that will use Artificial intelligence , computer vision technique more over this is customized solution for higher educational institute significantly that allows to monitor the daily reports, it allow us to generate alert message it allow us to detect contract tracing in one solution precisely, these are the different solutions which were different in every system there is no one unique software system that allows all these tasks to be performed together which is the scope significance of the proposed system.

Technical Details of Final Deliverable

The final deliverable of the project will be a desktop application along with documentation manual having all technical details of the application. It will be developed in

•  NumPy : Used for storing and manipulating high dimensional arrays.
• Matplotlib : Used for plotting.
• Scikit-Learn : Used for DBSCAN clustering.
• PIL : Used for manipulating images.
• OpenCV : Used for manipulating images and video streams.
• Keras : Used for designing and training the Face_Mask_Classifier model.
• face-detection : Used for detecting faces with Dual Shot Face Detector.
• face-recognition : Used for detecting facial landmarks.
• tqdm : Used for showing progress bars.
• Google Colab : Used as the developement environment for executing high-end computations on its backend GPUs/TPUs and for editing Jupyter Notebooks.

Anaconda will be use with Python for backend development.

Final Deliverable of the Project

HW/SW integrated system

Core Industry

IT

Other Industries

Education , Health

Core Technology

Artificial Intelligence(AI)

Other Technologies

Others

Sustainable Development Goals

Good Health and Well-Being for People

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com