Design And Implementation Of A Wearable Eye Tracking System For Affective Computing

Eye tracking is an experimental method of recording the position of the eyes,which can be used to observe the distribution of visual attention,and has been widely used in the fields of psychology,medicine,neuroscience and computer science.With the continuous development of eye tracking technology,more and more researchers have paid more and more attention to this technology.Eye movement has become one of the most commonly used features in affective computing because it can pinpoint the gaze area of users,characterize their subconscious behaviour,and provide some emotion-related features to determine the user’s emotional states.With the widespread use of eye tracking in affective computing,related hardware devices and software systems have also developed rapidly.There are two main types of eye trackers in common use today: desktop eye trackers and wearable eye trackers.Compared to desktop eye trackers,wearable eye trackers are more suitable for near-eye eye tracking related research and can extract clearer eye data.However,the existing wearable eye trackers has a complex structure and relatively large shielding of the face,which affects the experience of the subjects during the experiment.The software system also has problems,such as the inability to collect multimodal data,the low robustness of the pupil detection algorithm,which affects the accuracy of fixation prediction,and the fact that it is not fully suitable for research related to emotional computing.In addition,some softwares do not have a feature extraction function or the function is not perfect.It is not friendly for researchers in non-computer-related fields such as psychology and medicine,and increases the time cost of feature extraction.In order to solve the above problems,this thesis designs and develops a wearable eye-tracking system for affective computing.The system has been improved and optimized in terms of hardware and software:(1)In terms of hardware,the structure is optimised to reduce facial occlusion,and infrared light sources and filters are added to adjust the light to eliminate the effect of ambient light on eye movement data collection.(2)In terms of software,a hard real-time synchronisation scheme is proposed to realise synchronous acquisition of eye movement and other modal data,and a highly robust pupil detection algorithm is proposed to improve the accuracy of fixation prediction.At the same time,an eye movement feature extraction tool module is also designed and implemented,making it more suitable for affective computing researchers.The main research content of this thesis is as follows:(1)The wearable eye tracker designed in this thesis is optimised for the complex structure of the eye tracker on the market and the large occlusion of the face.Its hardware structure has been optimised to minimise the occlusion of the subject’s face.Apart from the eye-tracking camera,the eye-tracking device does not contain any other structures that obscure the face,which improves the subject’s level of comfort during the experiment.At the same time,the eye tracker retains important facial information that can be used for multimodal research on expression and eye movement.To provide a good data collection environment for the software,the wearable eye tracker is equipped with an infrared light source and a filter to adjust the light,eliminating the influence of ambient light on eye movement data collection and improving the clarity of eye tracking camera’s capture of the eye movement.(2)In order to solve the problem that the software system cannot carry out multimodal data acquisition and the low robustness of the pupil detection algorithm affects the accuracy of fixation prediction,this thesis designs and develops the host computer software system to match the hardware.Aiming at the problem that some softwares in the market cannot collect multimodal data,this thesis uses the software in a hard realtime operating system to optimise the multimodal synchronous acquisition process and greatly optimise the data frame alignment,thereby improving the quality of the original data.Even if it runs on a low-performance computer or a high-load computer,the data acquisition process can maintain stability and realise multimodal synchronous data acquisition.After testing,when the computer is under heavy load,it can still ensure that the maximum process delay difference is within 1.2ms,and the maximum process delay difference of ordinary operating systems is as high as 120 ms.Aiming at the problem that the low accuracy of pupil detection in some software systems affects the accuracy of gaze point prediction,this thesis proposes a high-precision pupil detection algorithm,which uses adaptive threshold segmentation and threshold noise removal to improve the robustness of the algorithm,and can avoid the influence of irrelevant features such as eyebrows on pupil detection.In addition,this thesis also uses a regression-based twodimensional gaze point prediction model to realise fixation prediction.The algorithm model is simple and the computation speed is fast,which is suitable for project requirements.(3)In order to solve the problem that some software is lacking or the feature extraction function is not perfect,and it is not friendly for non-computer professional researchers(such as psychology,medicine,etc.)to conduct research related to emotional computing,this thesis designs and develops an emotion-oriented computational eye movement feature extraction module.This functional module can conveniently extract eye movement related features and data indicators.It provides more feature types and dimensions than existing feature extraction software in the market,and can display some indicators visually.The software has a simple user interface and good human-computer interaction,which greatly saves the time cost of feature extraction for researchers.