谢子翰,顾宏斌,吴东苏.基于多目摄像机的鲁棒眼动跟踪技术研究[J].测控技术,2022,41(12):58-65 |
基于多目摄像机的鲁棒眼动跟踪技术研究 |
Robust Eye Tracking Technology Based on Multi-Camera |
|
DOI:10.19708/j.ckjs.2022.02.224 |
中文关键词: 眼动跟踪 头部姿态评估 加权融合 多目摄像机 |
英文关键词:eye tracking head posture assessment weighted fusion multi-camera |
基金项目: |
|
摘要点击次数: 352 |
全文下载次数: 902 |
中文摘要: |
为改进单个远程传感器采集眼动数据时存在视场小、容易被遮挡物遮挡的缺点,研究了一种多目摄像机眼动跟踪技术,以更好地采集眼动数据。应用多个摄像机对人体头部姿态和眼球信息进行采集,通过分割视频帧提取瞳孔图像和计算被测用户头部姿态角度,将瞳孔图像放入卷积神经网络进行训练得到注视点坐标,并基于头部姿态信息计算每个摄像机的注视点权重,从而加权融合得到更精确的注视点信息。研究结果表明:在头部姿态角较大时,多目眼动追踪技术的精度比单目传感器的精度高30%~50%。该技术具有灵活性和通用性,在驾驶舱设计、用户用眼习惯评估和驾驶学员眼动绩效分析中具有重要的应用和推广价值。 |
英文摘要: |
To improve the shortcomings of a small field of view and easy to be blocked by obstructions when a single remote sensor collects eye movement data,a multi-camera eye-tracking technology is studied to better collect eye movement data.Multiple cameras are used to collect human head posture and eyeball information.Through extracting the pupil images by segmenting video frames and calculating the head posture angle of the measured user,the pupil images are put into the convolutional neural network for training to obtain the fixation point coordinates,and the weight of each cameras gaze point is calculated based on the head posture information to obtain more accurate gaze point information by weighted fusion.The research results show that the accuracy of multi-eye eye-tracking technology is 30%~50% higher than that of monocular sensors when the head posture angle is large.This technology is flexible and universal.It has important application and promotion value in cockpit design,user eye habit evaluation and driver student eye movement performance analysis. |
查看全文 查看/发表评论 下载PDF阅读器 |
关闭 |