The Face Recognition Web Application for Crime Prevention is an application that utilizes machine learning technology to recognize faces and prevent crime.
The system is designed to identify suspects by comparing their facial features with those in a database of known criminals.
Supporting Material: https://drive.google.com/file/d/1pcG-72WVB-8UwsnUoz_L73FGz-k1Bvim/view?usp=sharing
The research was conducted from May 2022 - July 2022 and was associated with the Indian Journal of Applied Research (IJAR).
The research findings have been accepted for publication in the 1st September 2022 issue of the IJAR.
01) Machine Learning Model: is trained using labeled images and is capable of recognizing faces with 98%+ accuracy. Once the model is trained, it is exported as a JSON file and linked with the web application.
02) Web Application: is designed to work with the deployed machine learning model and can detect faces with 98%+ accuracy within 4 seconds. The application uses a camera to capture images of people in a given area, which are then processed by the machine learning model to recognize the faces.
Whenever a person is recognized with 98%+ accuracy, the web application sends an email to the admin and/or the police. This allows for quick identification and intervention in case of suspicious or criminal activity.
The technology behind this application can be an effective tool for crime prevention and can be used in various settings such as public places, residential areas, and businesses to enhance security and safety. The application can also be customized to fit the specific needs of different users, making it a versatile solution for preventing crime.
On executing file face_recognition_model.ipynb in the Github repository, the following steps are done:
- Capture images to train a model.
- Image Labelling of captured images
- Partitioning of labeled images in the Training & Testing section
- Model Training
- Model Evaluation
- Model Export (JSON file is created)
- Python
- Tensorflow Object Detection API
- OpenCV
Then we host this JSON File and link it with our web application. The web application uses the deployed Machine Learning model and can detect faces with 98%+ accuracy within 4 sec.
- Backend: Next.JS, Node.JS
- Database: MongoDB
- Frontend: HTML5, CSS3, Javascript & JSX
- Version Control: Git
Google Colab Notebook Link: https://colab.research.google.com/drive/1NeGuc0xFChaScWOOub5GbGsoudWhNdH1?usp=sharing
Step 1. Open Notebook: https://colab.research.google.com/drive/1NeGuc0xFChaScWOOub5GbGsoudWhNdH1?usp=sharing
Step 2. Set Notebook settings to GPU (It is required for training model)
Step 3. Run cells upto folder setup
Step 4: At section 1.4 Capturing Images you can uncommment the code & run it to capture at that moment or you can run next cell to add images in Tensorflow/workspace/models/my_ssd_mobnet
Step 5: After section 1.5 Labeling Images you need to manually divide collected images into two folders train and test. So now all folders and annotations should be split between the following two folders. \Tensorflow\workspace\images\train \Tensorflow\workspace\images\test
Step 6: At section 2.11 Load Train Model From Checkpoint: Please make sure to enter latest Checkoint name and then run this cell. You can find it in Tensorflow/workspace/models/my_ssd_mobnet
Step 7: Run all cells
After excecuting all the cells you will get your model as model.json in Tensorflow/workspace/models/my_ssd_mobnetTensorflow/workspace/models/my_ssd_mobnet/tfjsexport
Step 8: You can optionally evaluate your model inside of Tensorboard.
Once the model has been trained and you have run the evaluation command under
Navigate to the evaluation folder for your trained model e.g. cd Tensorlfow/workspace/models/my_ssd_mobnet/eval and open Tensorboard with the following command
tensorboard --logdir=.
Tensorboard will be accessible through your browser and you will be able to see metrics including mAP - mean Average Precision, and Recall.
- System requirements:
a. Windows: Windows 7 or higher (64-bit)
b. Linux: If Ubuntu 16.04 or higher (64-bit)
-
Hardware requirements: NVIDIA GPU card with CUDA architectures 3.5, 5.0, 6.0, 7.0, 7.5, 8.0 and higher.
-
Software requirements: a. Python 3.7–3.10
b. pip version 19.0
c. Windows: Microsoft Visual C++ Redistributable for Visual Studio 2015, 2017 and 2019
d. NVIDIA® GPU drivers version 450.80.02 or higher.
e. CUDA® Toolkit 11.2.
f. cuDNN SDK 8.1.0.
Step 1. Clone this repository: https://github.com/apoorva-01/Face-Recognition-WebApp-For-Crime-Prevention
Step 2. Create a new virtual environment
python -m venv frme
Step 3. Activate your virtual environment
source tfod/bin/activate # Linux
.\tfod\Scripts\activate # Windows
Step 4. Install dependencies and add virtual environment to the Python Kernel
python -m pip install --upgrade pip
pip install ipykernel
python -m ipykernel install --user --name=frme
pip install jupyter
Step 5. Run Notebook (Ensure you change the kernel to the virtual environment as shown below)
Step 6. Run cells upto folder setup
Step 7: At section 1.4 Capturing Images you can uncommment the code & run it to capture at that moment or you can run next cell to add images in Tensorflow/workspace/models/my_ssd_mobnet
Step 8: After section 1.5 Labeling Images you need to manually divide collected images into two folders train and test. So now all folders and annotations should be split between the following two folders. \Tensorflow\workspace\images\train \Tensorflow\workspace\images\test
Step 9: At section 2.11 Load Train Model From Checkpoint: Please make sure to enter latest Checkoint name and then run this cell. You can find it in Tensorflow/workspace/models/my_ssd_mobnet
Step 10: Run all cells