-
-
Advait keskar’s face being recognised by the module
-
Part of our code
Inspiration
In our day to day college life we noticed a major flaw in college operations, viz. Physical Attendance. Day in and day out our professors keep telling us to maintain 75% or more attendance, but the way in which colleges store or record attendance is primitive. And obviously students can easily trick this system by proxy attendance. So we came up with an idea of creating a facial recognition system which can recognize a student within milliseconds. A system which can log their in and out time and keep a meticulously maintained excel file. This system will greatly reduce the workload of Professors and give students 10-15 extra minutes to study or interact with professors.
What it does
This system has been divided into two parts : Trainer and Recognizer. The Trainer takes prestored images of students and converts these images into grayscale and then into NumPy array and stores the arrays into a .YML file and the Names associated to these students and additional data into a .pickle file.
The Recognizer takes live images from the video capture system, creates a rectangular region of Interest, converts the image within the ROI to grayscale and then to a NumPy array. This live NumPy array is then compared to prestored arrays and a confidence value is spit out. According to this confidence value we recognize the student.
Challenges we ran into
Our very first challenge was to detect Multiple faces at once, we are currently working on this problem. Honestly this problem is a trivial one if we use third-party pre defined python modules but we have made the whole recognizer from scratch and we take great pride in that.
Next Challenge was what if a student hides his face while leaving the class and effectively bunking it, for that we have added a csv function which detects whether a attendance record is valid or not.
Next was what if the system becomes slow, we are not worried about it as the effective scantime for eerie face is approximately 14 milliseconds.
Accomplishments that we're proud of
We have managed to create a Facial Recognition system which has a 14 ms per scan. We have made the whole project from scratch without using any third-party predefined library. We have managed to create and maintain a csv file for records
What we learned
We learnt new concepts about Machine Learning and Computer Vision. We learnt that small innovations and a bit of initial effort can greatly reduce a professor’s workload. We learnt concepts of real time data analysis.
What's next for Attendance using Facial recognition Team-81
For Attendance system for facial recognition, the future is to create a camera module that can be installed on a doorframe which can scan and detect faces. This system can be greatly optimized and it can be effectively attached with an ERP system to automate the process of attendance and record keeping.
Log in or sign up for Devpost to join the conversation.