A method and system for saccade signal recognition based on fusion of bimodal classification models

Abstract

The invention discloses a saccade signal recognition method and system based on dual-modal classification model fusion, which belongs to the technical field of electrooculogram. The method includes: synchronously collecting EOG data and video data of different saccade action categories of subjects; and video data are preprocessed separately; endpoint detection is performed on EOG data and video data; the endpoint with longer valid data in the EOG data endpoint detection result and the video data endpoint detection result is selected as the final endpoint detection result; The effective eye movement data segment is divided into a training set and a test set, and the feature parameters are extracted; the feature parameters of the effective eye movement data in the two modes are sent to the SVM classifier for training, and two classification models are obtained; the two classification models are Fusion; model fusion is tested using data from the test set to recognize glance signals. The fused features in the present invention have more complementary information, which improves the robustness of signal recognition.

Description

technical field [0001] The invention relates to the technical field of electrooculograms, in particular to a method and system for identifying glance signals based on fusion of dual-mode classification models. Background technique [0002] Human Activity Recognition (HAR) is the recognition and representation of individual behaviors, interactions between people and between people and objects. At present, it has been widely used in key research fields such as motion analysis, virtual reality and patient monitoring. [0003] Among them, eye movement is a relatively common activity in daily behavior activities, and plays an important role in human-computer interaction, cognition, drug effects, and psychology. In the EOG-based HAR system, the recognition of glance signals plays an important role in the final human action recognition results. In order to realize the effective recognition of the saccade signal, the research scheme proposed so far is mainly as follows: [0004] ...

AI Technical Summary

Problems solved by technology

[0007] However, the current EOG identification has the following problems: First, strict EOG acquisition conditions are required. In some environments with relatively large noises, it is difficult for EOG to accurately describe the characteristics of the original signal; second, in the EOG acquisition process It is necessary to use multiple leads for data collection in order to obtain rich eye movement information, but the interaction between multiple leads will also bring bias to the final human behavior recognition

[0008] In contrast, the video-based HAR system can well overcome the above problems, but the recognition process based on video data is easily interfered by optical fibers, and the performance of the video-based HAR system will drop sharply in poor light environments

[0009] Therefore, the accuracy of the existing single-modal glance signal recognition results is not high, and it is difficult to apply to changing environments

Images