Using Convolutional Neural Networks with External Webcams to Develop
a Reliable and Economical Webcam-Based Eye Tracker
Tracking eye movement is particularly helpful in gaining insight into cognitive health.Eye Trackers (ETs) are advanced, high-resolution, external cameras that follow and measure eyemovement. Studies that use ETs found that irregular ocular saccadic paradigms (such as slowerand less accurate saccades) have strong correlations with neurodegenerative diseases (NDs).Treatments for NDs can be more effective at earlier stages, making early diagnosis of cognitiveimpairment (CI) crucial. ETs could be used to provide a provisional diagnosis of CI; however,their high cost, upwards of $40,000 excluding software, limits accessibility and feasibility ofprogress in this area. In this research, an image-classification convolutional neural network(CNN) was investigated using images captured from 3 external webcams to mimic an ET withsimilar reliability. The model calibrated itself by displaying a dot and asking the user to followthe dot as it moved. The webcam captured images as the dot moved and recorded the position ofthe dot every time it captured an image. Approximately 750 images of the eyes for each webcamwere captured; 76% of these images were used to train the model while 24% of the images wereused for validation. Accuracy levels achieved using the CNN webcam-based ET werecomparable to existing technologies, with errors typically ranging from 1.77% to 3.83%. Futuresteps include experimentation within a clinical setting in order to understand the efficacy of themodel in providing a cognitive health diagnosis based on saccadic paradigms.
Akshat enjoys participating in math competitions, conducting research, and serves as the president of the computer science club. He is a member of the Mu Alpha Theta Honor Society and the Science National Honor Society. After high school, Akshat wants to pursue a degree in computer science before becoming a software engineer.