Lava Kumar
Lava Kumar has a total of 5 co-authors. Here are the top co-authors by number of shared publications:
- Bharat Singh Shekhawat
Entropik Technologies Pvt. Ltd., Bangalore 560017., India
1 shared publication - Gayathri Sowmya G
Entropik Technologies Pvt. Ltd., Chennai 600032., India
1 shared publication - S. J. Dharbaneshwer
Department of Engineering Design, Indian Institute of Technology, Madras, Chennai, India
1 shared publication - Soumitra Ghosh
Entropik Technologies Pvt. Ltd., Bangalore 560017., India
1 shared publication - Sumit Singh Chauhan
Entropik Technologies Pvt. Ltd., Bangalore 560017., India
1 shared publication
Latest Publications
- Open Access
- 4 January 2024
- Published by Association for Computing Machinery (ACM)
- p. 298-306
Real-time eye tracking using representation learning and regression
Eye gaze serves as a primary indicator in identifying the human interest and attention for human-computer interaction (HCI) studies. Most of the eye tracking technologies uses either specialized hardware or requires large number of calibration images to provide accurate results. Although real-time eye tracking approaches are proposed in the literature, most of these approaches use eye attributes such as pupil center and eye-lid for gaze estimation. These features are not robust, sensitive to even minor illumination changes and varies from person-to-person as well. In this work, we propose a novel representation-learning based eye tracking framework that works real-time and requires only limited number of calibration images. The person-to-person variations in eye appearance such as eye lid size and foveal offset is taken into consideration via a regression model. To account for varying head pose during eye tracking, compensation mechanisms are integrated to our pipeline. In order to illustrate the efficacy of our framework, point-wise validation and pursuit-based validation are conducted, and the gaze estimates are found to be accurate. The system is also compared with a baseline model such as Webgazer using open-source benchmark dataset- Eye of the typer Dataset (EOTT). To showcase the usability of our framework, the gaze behaviour of users on web pages and products displayed in shelf are analyzed using our system. It is observed that the gaze analysis results obtained are reliable, intuitive and aligned with the human insights.
- 4 January 2024
- Published by Association for Computing Machinery (ACM)
- p. 298-306
Open AccessReal-time eye tracking using representation learning and regression
Eye gaze serves as a primary indicator in identifying the human interest and attention for human-computer interaction (HCI) studies. Most of the eye tracking technologies uses either specialized hardware or requires large number of calibration images to provide accurate results. Although real-time eye tracking approaches are proposed in the literature, most of these approaches use eye attributes such as pupil center and eye-lid for gaze estimation. These features are not robust, sensitive to even minor illumination changes and varies from person-to-person as well. In this work, we propose a novel representation-learning based eye tracking framework that works real-time and requires only limited number of calibration images. The person-to-person variations in eye appearance such as eye lid size and foveal offset is taken into consideration via a regression model. To account for varying head pose during eye tracking, compensation mechanisms are integrated to our pipeline. In order to illustrate the efficacy of our framework, point-wise validation and pursuit-based validation are conducted, and the gaze estimates are found to be accurate. The system is also compared with a baseline model such as Webgazer using open-source benchmark dataset- Eye of the typer Dataset (EOTT). To showcase the usability of our framework, the gaze behaviour of users on web pages and products displayed in shelf are analyzed using our system. It is observed that the gaze analysis results obtained are reliable, intuitive and aligned with the human insights.