Improving Eye Movement Biometrics Using Remote Registration of Eye Blinking Patterns

随着先进技术越来越多地融入我们的日常生活，为了提高安全性和隐私性，并减少规避风险，对可靠的生物识别认证和身份识别系统的需求比以往任何时候都更加明显。尽管精度很高，但目前的生物识别技术依赖于特定的物理特征，如指纹、面部和虹膜特征，很容易受到欺骗和伪造。因此，确实需要开发更可靠的人体识别生物识别解决方案，能够在不需要特定动作（例如将手指放在扫描仪上或查看虹膜相机）的情况下持续识别/验证人。在最近的NSERC Engage项目中，我们**进行了一项基于眼部运动和眨眼的远程视频捕获的人类识别的可行性研究。通过融合眼球运动（凝视）和眨眼两种生物识别模式，我们能够**获得非常好的结果，识别率高达95.83%，错误率高达2.15%。数据是在多伦多大学实验室收集的，来自22名志愿者观看精心挑选的视频片段。在这个**项目中，我们将进行同样的实验，将摄像头安装在汽车仪表板的前部，面向驾驶员。由于有不同的（和不断变化的）照明条件和更多的运动由**司机完成，可能会有更多的异常值和变化的数据，可能会改变结果。此外，模拟的驾驶结果将与真实的驾驶结果进行对比，从而将理论与实践相结合。收集到的数据还将决定该系统在非侵入性方面的实用程度，以及它能够**连续验证驾驶员身份的频率——尤其是在极端情况下。为了完成这项任务，我们计划**使用我们的工业合作伙伴ACS公司提供的真实驾驶模拟器进行实验。

With increasing integration of advanced technologies into our daily life, the need of reliable biometric**authentication and identification systems is more pronounced than ever before in order to improve security and**privacy and to reduce the risk of circumvention. Despite high accuracy, current biometric technologies relying**on particular physical features such as fingerprints, face and iris characteristics are vulnerable to spoofing and**forgery. Therefore, there is a real need for development of more reliable biometric solutions for human**recognition that are able to continuously identify/authenticate people without requiring specific actions from**them (e.g. placing fingers on an scanner or looking into an iris camera). In a recent NSERC Engage project, we**carried out a feasibility study on human recognition based on remote video capture of eye movements and**blinking. By fusing two biometric modalities that is eye movement (gaze) and eye blinking we were able to**achieve very good results that is up to 95.83% recognition rate and 2.15% error rate. Data were collected in the**University of Toronto Laboratory from 22 volunteers watching carefully selected video sequences. In this**project we will conduct the same experiment with a camera attached to the front of a car's dashboard, facing**towards the driver. As there are different (and changing) lighting conditions and more movements done by the**driver, there may be more outliers and variation in the data that may alter the results. Furthermore, it will test**the theory with practice as the simulated driving results will be put up against a real session. The data collected**would also determine how pragmatic the system is in being non-invasive, as well as how often it can**continuously authenticate the driver- especially in extreme cases. In order to accomplish this task, we plan to**perform experiments using the real life driving simulator available at our industrial partner, ACS corp.