A YOLO Detector Providing Fast and Accurate Pupil Center Estimation using Regions Surrounding a Pupil

Abstract

Eye-tracking technology has many useful applications, including Virtual Reality (VR) devices, Augmented Reality (AR) devices, and assistive technology. The main objective of eye-tracking technology is to detect eye position and track eye movements. It is possible to determine the eye position when the pupil center is detected. In this paper, a deep learning-based approach to the detection of pupil centers from webcam images is presented. As opposed to all previous approaches to object detection based on training the detector with objects to be detected, our object detector was trained with both the region surrounding a pupil and the region between an eye and the region surrounding a pupil. The latter set of regions has been found to increase the overall detection accuracy. A novel post-processing algorithm is also presented to estimate the pupil center from all the detected regions. To achieve real-time performance, we have adopted the tiny architecture of YOLOv3, which has 23 layers and can be executed without requiring a GPU accelerator. To train the detectors, different variations of regions covering a pupil and an eye were used, as well as expanding regions surrounding a pupil and an eye. The PUPPIE dataset was used as the primary input for training the detector. The setting with the best detection accuracy was applied to all publicly available datasets: I2Head, MPIIGaze, and U2Eyes. In terms of accuracy, the results indicate that pupil center estimation is comparable to the state-of-the-art approach. It achieves pupil center estimation errors below the size of a constricted pupil in more than 98.24% of images. Furthermore, the detection time is 2.8 times faster than the state-of-the-art approach. Doi: 10.28991/ESJ-2022-06-05-05 Full Text: PDF

Fast and Accurate Pupil Estimation Through Semantic Segmentation Fine-Tuning on a Shallow Convolutional Backbone

“…Larumbe-Bergera et al [14] and Kurdthongmee et al [15] compared deep learning methods with other advanced approaches, underscoring improvements in both accuracy and computational efficiency. While these methods marked an improvement over previous models, they still face challenges in balancing accuracy with processing speed, particularly in real-time scenarios.…”

An analysis of precision: occlusion and perspective geometry’s role in 6D pose estimation

“…Similarly, CNNs have been shown to be able to identify key points with their own descriptors that are more resilient to image changes than the legacy methods and have extremely low latency [10,11,21,25]. However, all of these key point identification methods thus far hide information inside their respective algorithms with custom descriptors for features that are found, which cause difficulty if used for 2D-3D correspondences.…”

The Eye: A Light Weight Mobile Application for Visually Challenged People Using Improved YOLOv5l Algorithm

“…Mobile applications designed for visually challenged people enhance their accessibility, object recognition [8,9], social interaction and communication, education and learning, independent living etc. The YOLO (You Only Look Once) object detection algorithm [3,10,11] has evolved, including YOLOv4 [12] and YOLOv5. YOLOv4 improved accuracy and introduced architectural enhancements, while YOLOv5 simplified the architecture and prioritized real-time inference.…”

“…They used the YOLOv4 methodology to detect vehicles and enhanced the methodology's speed without sacrificing its accuracy. The real-time object detection method is based on improved YOLOv4-tiny by Jiang et al [11]. The author used an improved YOLOv4 methodology to improve the speed of real-time object detection.…”