Mobile Device Eye Tracking Data Acquisition System Based On Crowdsourcing

Eye tracking has become a promising research area,and is wide utilized recently.However,due to limited hardware performance and computing resource,eye tracking is hard to achieve high accuracy and efficiency on mobile device.It often needs the support of external hardware.but the cost of development increases and is harder to use.For this purpose,this paper proposed a method to ask users receall the gaze and report them based on the mechanism of crowdsourcing,and developed an eye tracking data acquisition system for mobile devices.The specific research work includes the following three aspects:(1)Crowdsourcing method for user gaze-recall task.Under the environment of context awareness,crowdsourcing technology was used distribute gaze-recall task and aggregate task results.Task publishers used the crowdsourcing platform to publish gaze-recall tasks,and got the task results with viewing the result visualization.Workers could query,accept and complete the gaze-recall task on the crowdsourcing platform.(2)Task recommendation based on context aware.In order to improve the calculation accuracy of user gaze recall data,and reduce the noise data,context aware technology based on location and velocity context was used to make recommendation of gaze-recall tasks.According to the user’s current location,such as in the indoor or outdoor,suitable tasks was recommended to the user,with increasing or decreasing the number of test images appropriately;simlarly,according to user’s movement speed,tasks with different difficulty were recommended to the users.Compared the gaze-recall data by using context aware technology and without context aware technology,the result showed that,after using context aware technology,the efficiency of task completation was improved with average 1-1.3 seconds per test image,and the accuracy of gaze-recall data was improved with 25% to 30%.(3)Error compensation model based on support vector regression.In order to further reducing the noise data,this paper mapped the relationship between the fixation data acquired by gaze-recall and acquired by the eye tracker using the support vector regression method.And then calibration method was used to train the error compensation model for refineing the gaze-recall data.The error compsensation model could reduce the diffirence between the real fixation data and the gaze-recall data,and improve the accuracy of the gaze-recall data.Compared the gaze-recall data by using the error compensation model and without the error compensation model,the results showed that,after using the error compensation model,the accuracy of the gaze-recall data is improved,and for different types of tasks and test images,the accuracy was improved with 15% to 40%.Based on the research mentioned above,a prototype system for eye tracking data acquisition based on smart phone was bulit.Then the gaze-recall data visualization and analysis was proposed,with drawing the hot map and the probability density distribution map.Compared to the fixation data acquired by the eye tracker through the user study,the Chi-square distance between them was 0.3 to 1.5,and the Pearson correlation coefficient was 0.4 to 1.The results showed that the data collected by gaze-recall achieved a high consistency with the real eye tracking data,and this validated the feasibility and effectiveness of the eye tracking data acquisition method in this paper.