Eye Contact During Live Video Conferencing Using External Video Camera

Video conferencing and other forms of live communication methods enjoy widespread popularity. However, the absence of interpersonal, psychological, or tactile feedback diminishes engagement compared to in-person communication. A notable challenge arising from the distance between the webcam and display of the device lies in the absence of contact of eye during live video conferences. Addressing this issue involves manipulating video frames to alter the direction of the gaze, creating the illusion of sustained eye contact. This research introduces a technique for preserving eye contact, entailing the rotation of a user's eyeball prior to transmitting the frame of video to the distant place. The implementation leverages WebRTC and TFLite within a application optimized for mobile devices. Convolutional neural networks based on warping play a pivotal role in superimposing the corrected eye onto the detected eye position during the eye position correction process. The employed loss functions contribute to maintaining the geometry of the eye structures while mitigating color shifts around the borders of the eye regions. The technology's real-time eye gaze rectification capability on consumer-oriented mobile devices is demonstrated, highlighting its remarkable speed and resolution. This approach to gaze correction, achieved through postprocessing, proves to be a practical solution for addressing the challenge of lost contact of eye during real time video communication. The model's integration into an application facilitates its implementation.