IIIT-B develops webcam-based tool to track distraction in surveys

The International Institute of Information Technology Bangalore (IIIT-B) is developing a low-cost eye-tracking tool that uses a regular webcam to detect when people answering long surveys become distracted or mentally fatigued, a step to improve the reliability of public health and social survey data.

The project is funded by Machine Intelligence and Robotics CoE (MINRO).

The large-scale health and behavioural surveys are mostly carried out through door-to-door visits, with interviewers use questionnaires to ask respondents. These surveys are important for shaping public policy but are expensive and demanding to conduct. During the Covid-19 pandemic, such in-person surveys became difficult, pushing agencies to rely more on online surveys. However, researchers found that self-administered surveys often have high dropout rates and incomplete answers, particularly when respondents feel overwhelmed.

The project, led by IIIT-B professor Jaya Sreevalsan Nair with Beryl Gnanaraj, a PhD candidate at the institute, addresses this gap. The team explored whether eye-tracking could help understand respondent behaviour during surveys. The idea is to identify signs of cognitive overload such as loss of focus or prolonged hesitation which often lead to skipped questions or unreliable responses. By flagging such moments, survey organisers can better assess the quality of the data or redesign surveys to reduce respondent fatigue, Prof. Nair said.

Beyond healthcare surveys, the researchers believe the technology could be adapted for other uses including assessing reading ability in children with learning disabilities, monitoring attention in digital learning platforms, identifying malpractices during online examinations, and exploring applications in mental health research.

Currently, eye-tracking solutions rely on specialised hardware that is largely developed in the United States or Europe and is extremely expensive. According to the researchers, a professional eye-tracking device, along with its supporting software, can cost close to ₹50 lakh. This cost barrier was a key reason for designing a solution that works with standard webcams.

The system uses webcam footage to generate visual maps showing where a respondent is likely looking on the screen. These visualisations help identify gaze points, which are then analysed using computer models to estimate mental effort and attention levels. The goal is to understand whether a respondent is focused, distracted or struggling with a particular question, without requiring any additional equipment.

To make this possible, the researchers are using advanced machine learning techniques, including deep learning models, to estimate gaze direction from webcam images. Unlike professional eye trackers that directly capture precise eye movements, webcam-based systems must work with noisier data. To improve reliability, the IIIT-B model uses both raw webcam footage and processed visual cues, a method the team says sets it apart from most existing webcam-based trackers.

The system is currently designed for post-survey analysis, meaning the webcam footage is reviewed after the survey is completed. While real-time analysis is not yet part of the project, it may be explored in future stages. At present, the team has conducted qualitative assessments using visual outputs and found the results encouraging, though a formal accuracy comparison with professional eye-tracking devices is yet to be done.

The tool has not yet been piloted in live survey settings. The team is in discussions with the National Institute of Mental Health and Neurosciences (NIMHANS), Bengaluru, to first use the system in a research environment before testing it in real-world survey conditions.

One of the major challenges in developing the tool has been creating a large training dataset. Annotating video data frame by frame is time-consuming, and building enough examples to reflect real-world conditions such as poor lighting, head movement, different facial angles and low-resolution webcams has required significant effort.

The researchers have also emphasised the importance of privacy, given that the system involves recording facial video and eye-gaze information. The project is being reviewed by the Institute Review Board, and the data is currently being used only for research purposes. Any future data sharing or publication will comply with applicable legal and ethical guidelines.