When I speak of face recognition, people often refer to seamless facial recognition systems seen in airport immigration counters, renowned for their efficiency and accuracy. That is true and fair. What is more complex is implementing face recognition in not-so-controlled environments and for use cases that are far and wide. For example, the deployment of face recognition verification of voters in elections presents a uniquely complex set of challenges compared to the controlled environment of airports. What is more complex than elections in India!
In airports, specialized cameras with height and light adaptation capabilities are paired with high-end hardware, meticulously prepared for optimal face recognition. Lighting conditions are carefully controlled, ensuring even illumination without back-lighting. Moreover, reference photos adhere to standards, typically recent and of high quality.
Contrastingly, deploying facial recognition in elections entails unparalleled obstacles. SvaDESH relies on basic Android mobile cameras, which may struggle to capture accurate facial images in uncontrolled environments with variable lighting conditions. Reference photos often comprise aged, low-resolution images, some even in black and white, posing significant hurdles to accurate identification. Additionally, the age gap between reference photos and voters can span decades.
In terms of timeline, elections demand a rapid turnaround, with deployment requiring just 12 hours compared to a minimum of 3-6 months seen in airports. Training personnel for election booths is condensed into 1-2 days, with a lower level of technical expertise compared to airport staff.
Moreover, while airport passengers are educated and prepared for facial recognition, voters in elections may be unfamiliar with the process, encountering challenges such as wearing head coverings but have to identify and verify them in-spite of it.
While airport deployments benefit from controlled environments and extensive preparation, deploying facial recognition in elections demands adaptability, speed, and the ability to navigate myriad challenges within a constrained time frame. Despite these complexities, at FaceTagr we have figured how to navigate these challenges and have successfully deployed in 3850 booths with a voter count of 2.5M and have verified with very high accuracy.
I am very proud of the team.