Attendance Monitoring with Computer Vision

February 13, 2020

From self-driving cars like Tesla and Waymo to automated stores like Amazon Go, modern world is witnessing the era of computer vision (CV). How Facebook tags users in a group-photo or live video, makes us wonder. It uses facial recognition technology, a method of identifying human face. Let’s find out how this technology can be leveraged for implementation in the attendance monitoring system (AMS).

How does facial recognition work?

Facial recognition is a vast domain, so before going in depth, let’s understand this technology in a nutshell. Firstly, the picture of a person is captured from a photo or video, when appeared alone or in a crowd. The facial recognition software reads the geometry of the face and extract key features like distance between the eyes, distance from forehead to chin, color of hair etc. The software identifies key features, sets facial landmarks and generates facial signatures from the image captured. This facial signature being a mathematical formula is used for comparison with the database of known faces. The difference between the obtained facial signature and database images are calculated through different mathematical techniques and the nearest is chosen as an optimum result. Finally, the determination is made, and the output indicates whether the image exists in the database or not.

How to implement facial recognition in Attendance Monitoring?

The first step to develop a facial recognition-based attendance system is to create a database, which will store all employees’ images along with their information. Every time a new employee joins the organization, their image needs to be updated in the database as a single truth of attendance monitoring data. In reality, CCTV footage has maximum chance of getting distorted images, so one needs to have a good collection of images stored in the database that will be used for training the machine-learning model as well as for the final testing purpose.

Once the database is created, our objective is to check the availability of the image obtained from CCTV footage in our database. The individual face extracted from the footage now needs to be processed. In our day-to-day life we identify individuals from their hair color, color of the iris, shape of the nose etc. These are the features of human face which are well identified by the human brain. . The Convolutional Neural Network (CNN) which is a deep, feed-forward artificial neural network, is used to perform the feature extraction of the face and then process the image and finally classify it so that one image can be distinguished from another.

Computer can visualize an image as an array of pixels, which depends on the image resolution. To obtain the features, convolution of the image is done with other filters so that the image blur level, sharpen level can be noted. The images can be too large after convolution; so to reduce their dimensions, pooling operation is performed which reduces the parameters of the image but does retain important information which will be later used do classify the image of every individual employee. The output obtained is optimized to arrive at the final output. Later, the generated output needs to be checked with the stored images to identify the employee and log his/her attendance. This, in essence, summarizes the working of attendance system using facial recognition.

Why switch to Computer Vision?

Using CV-based AMS will reduce the cost of traditional biometric systems, punching machines, and procurement and maintenance of smart card. The only thing required is CCTV cameras, which are already available for security of the offices. Use of transfer learning from popular CV models like ImageNet shall reduce the model implementation and maintenance drastically. A low-cost CV reduces the chances of unwanted, unauthorized, proxy access or tailgating to office premises without human intervention, which is not possible using the legacy systems. It will reduce any option of duplicity for attendance monitoring as the employee's individual face marks his/her attendance. The system is simpler compared to age-old methods and saves time and effort in logging attendance thereby increasing productivity.

Computer Vision is one of the most trending areas of experimentation in most industries. The ease of implementation, maintenance and application of a CV-based AMS shall add agility to organizations and tighten security as well. The model faces challenges as images obtained from real time CCTV are mostly occluded, rotated or zoomed out faces, which would require image correction measures. Besides, it needs a huge collection of individual images. Amazon Go, for example, takes advantages of CV to identify customers, create real-time campaigns by assessing mood; automates end-to-end retail experience by tracking users and objects to enable self- checkout of products.

Puja Mitra is a member of the Innovation and Product Engineering - Analytics (IPE - Analytics) group within TCS Platform Solutions. She is having strong skills in Python, Data Science and Machine Learning. She is responsible for multiple product engineering and innovations, working as a business intelligence and machine learning engineer. She has been contributing to the development and implementation of predictive analytics and embedded analytics use cases for the home grown HRMS and Procurement platforms of TCS Platform Solutions. She holds an engineering degree in Electronics and Communication from West Bengal University of Technology (MAKAUT)