Summary
When a business local to Excellerent experienced a theft of valuable inventory at one of their sites, the AI team at Excellerent developed a solution. The security threat was real, the loss was real, the surveillance and controls were too few and the suspects too numerous.
Building security and maintenance personnel require access to the clients floor, and all locks would open with the same set of “keys” as provided to the client personnel, thus assets in a storage closet was vulnerable to off hours theft.
The client need
A way to securely store key assets while not prohibiting building security and maintenance personnel from doing their job. The solution, enhance access to the client area with a facial recognition-based door lock authentication system. The solution had to be highly reliable, flexible, and fast, while still inexpensive and state of the art. Utilizing a configurable facial recognition solution integrated with entry doors on people spaces and equipment closets, Excellerent has proven the ability to allow authorized individuals simple access, and unauthorized individuals denied.
The challenges
- Ensure access is granted and logged for all personnel that need to get on the floor at any hour
- Simplicity, reliability, low cost, and a system that registers each entrant on a cloud-based time stamped database.
- Develop a solution that overcomes potential limitations of Facial Recognition solutions such as sensitivity to pose variations, illumination variation, camera quality, etc.
The solution
Using a simple one-time registration process for authorized individuals, and utilization of open-source AI software, Excellerent built a solution whose cost profile features software, equipment, and administrative costs that any client can afford, while still providing state of the art security. Through implementation of the solution, any client can use the facial recognition solution tied to a the solenoid of the door lock system to allow or deny access to almost any door, thus allowing various areas within a facility to be secured.
Workflow
- Face detection and capture happens via webcam when a person enters facility.
- Every user detected by the webcam will be checked for compatibility with the database by comparing the user’s face encoding with the encodings of the faces from the trained dataset (The Euclidean distance concept is used).
- The best match is returned with corresponding name that will be displayed on the screen.
- The solution is integrated with door entry system to open the door lock and enable the authorized visitor to enter the room
- If no match is found, then Unknown will be displayed which helps to alert the security and automatically deny the visitor entry.
- Names will be pushed into csv file along with the time stamp. This is to keep track of the information of people who were detected and at what time.
Advantages
- Utilization of open-source AI software
- OpenCV (Open-Source Computer Vision Library) is an open-source computer vision and machine learning software library with thousands of optimized algorithms to detect and recognize faces
- Haar Cascade frontal face recognizer to detect the face from our webcam.
- Improved accuracy and speed
- Face-Recognition library built using the “dlib” library helps to recognize the face of a person with 99.38% Accuracy.
- Imutils – is a series of convenience functions to expedite OpenCV computing on the Raspberry Pi
- The power of the deep convolutional neural network (CNN) has been used to generate depth maps.
- Uses scoring technique to enumerate accuracy that can be customized.
- This system will not just detectthe person but also store the information of the person detected in a Microsoft Excel File for future use.
Features of the System
- Register a facial image and file with employee name
- Train the model with collected images.
- Whenever an employee access a door secured through this system, the system will upload the name of Employee and time stamp to CSV file.
- Customer configurable confidence score (admin can allow / disallow access based on the confidence score)
- Generate a report with the frequency number of the authorized person and unauthorized persons (labeled as unknown).
- Liveness detection- this feature is built with a combination of Convolutional Neural Networks and Computer Vision to detect between actual faces and fake faces in real-time environment. The image frame captured from webcam is passed over a pre-trained model. This model is trained on the depth map of images in the dataset. The depth map generation have been developed from a different CNN model.
- Improved accuracy – for the system to distinguish between similar faces, SVM algorithm with scikit-learn python library is used on top of the face recognition API.
Tools and Libraries
Python 3.9- for building the code.
OpenCV – for face detection used Haarcascade classifier for image capturing stage which is collecting employees faces for training phase.
Imutils – is a series of convenience functions to expedite OpenCV computing on the Raspberry Pi.
Face-recognition library- the algorithm gives High accuracy of (99.38%).
CNN based dlib’s state-of-the-art algorithm that is built with deep learning .
Streamlit – an open-source Python library that makes it easy to create and share beautiful, custom web apps for machine learning and data science.
The system runs on rasberrypi 4 model B computer and Raspbian GNU/Linux 10 (buster) operating system.
The result
The client can register as few or as many employees and/or building personnel as are necessary to access a space. The system is fast, reliable, low cost, and provides traceability. In addition, the solution has failsafe’s not shown above to ensure the safety of the employee base and the access needs for all.
