Research On Interaction Intention Recognition Based On Eye Movement In Virtual Reality Environment

With the continuous development of new-generation information technologies such as 3D graphics technology and artificial intelligence,virtual reality has been widely applied in the field of mechanical engineering in various aspects such as design,production,processing,assembly,training,etc.The human visual channel is an important way for people to obtain information,and eye-based interaction is fast and low-cost.Therefore,research on interaction intention recognition based on eye movement data in virtual reality environments is significant for implementing adaptive interaction interfaces.At the same time,due to the differences between virtual reality environments and traditional two-dimensional screen displays,the three-dimensional characteristics of spatial depth information make traditional interaction performance evaluation methods and fixation identification algorithms no longer applicable,posing new challenges to the research of interaction intention recognition based on eye movement data.This thesis focuses on solving the problems of interaction performance evaluation,fixation identification,and interaction intention recognition in three-dimensional virtual reality environments,and considers the impact of spatial depth changes in virtual reality environments.The main innovative achievements include:(1)Aiming at the problem of performance evaluation of different interaction modalities in virtual reality environments,a modified Fitts’ law adapted to three-dimensional environments is proposed,and explore the definition of its key parameters-target width and movement distance-in virtual reality environments.The effectiveness of the modified Fitts’ law is verified through experiments.This thesis compares and analyzes the interaction performance of eye-based interaction and controller-based interaction in virtual reality through a performance evaluation method based on the modified Fitts’ law.The gap between eye-based interaction and mainstream controller-based interaction in virtual reality environments has been clarified,especially for interaction with 2D near-field objects,where the user experience brought by eye-based interaction is not as good as that brought by controller-based interaction.But for 3D far-field objects with depth changes,the user experience brought by eye-based interaction is similar to that of controller-based interaction,and even the accuracy of eye-based interaction is slightly better than that of controller-based interaction.This conclusion provides theoretical support for selecting far-field interactions as interaction tasks in eye-based intention recognition research.(2)A velocity,dispersion&vergence-threshold identification algorithm(I-VDVT)is proposed to address the issue of the impact of spatial depth changes on the accuracy of fixation identification in virtual reality environments.Based on existing fixation identification algorithms,this algorithm introduces pupil distance change as the vergence threshold.It considers the opposite movement of both eyes caused by spatial depth changes in the fixation identification algorithm.Compared with traditional algorithms,it can obtain more accurate fixation coordinates.In addition,an outlier processing method is also proposed,which can effectively eliminate the influence of outliers on the spatial accuracy of identification results.(3)A interaction intention recognition model based on eye movement data was constructed for far-field interaction tasks in virtual reality environments,achieving the recognition of interaction intentions for selection tasks and teleportation tasks.This model is based on a feature set composed of hand-eye coordination-related features,fixation-related features,saccade-related features,and pupil-related features and is trained using the gradient boosting decision tree method.Compared with existing recognition models,this model has better recognition performance,especially the hand-eye coordination-related features,which play an important role in improving recognition performance.Finally,a teaching scenario of mechanical processing based on the intention recognition system was designed and implemented.Furthermore,the proposed fixation identification algorithm and intention recognition model were applied to verify the effectiveness of the intention recognition model in practical application scenarios.This study applies eye movement modality to virtual reality adaptive interaction interfaces,promoting the development of intelligent human-computer interaction interfaces in virtual reality environments and promoting the broader application of virtual reality technology in mechanical engineering.