Research On Saccadic Angle Recognition And Classification Of EOG Signal

The growing popularity of the computer facilitates more and more healthy people, while do not benefits those disabled individuals with server paralysis of the limbs, because they can not communicate with computers, even care themselves. Thus, HCI (Human-Computer-Interaction) systems based on bio-electrical come into existence. The biological signal of eye movements contains many features, which are easy detection, high amplitude, and uncomplicated treatment and so on, to make it popularization and to be a new kind of HCI.EOG (Electrooculogram, EOG) is becoming a new kind of acquisition method for eye movement's signal because of its non-invasive and little affect to normal activities. The EOG-based HCI system is also becoming a current focus of research, and it will have a more broad research prospect.This thesis mainly researches the feature extraction and pattern recognition of EOG signals under the different saccade angles. It proposes a set of effective parameters through the analysis of EOG signals, and makes the pattern recognition by using neural network. The main works are as follows:1. Experimental design and data collection based on saccadic angle position of EOG. According to the scope of the human saccade, we set four horizontal saccade angles initially, and then do a large number of experiments and record the original EOG data.2. Feature extraction of EOG signals. After reading a large number of literatures and continuously doing simulation experiments, we propose two kinds of feature parameters according to different saccade angles in this thesis:One is four-dimensional parameter based on the waveform, another is 15-dimensional combination parameter based on LPC and the main waveform characteristics.3. Pattern classification of the four saccade angles. To identify the four classes of modes, the thesis adopts BP network, RBF network and SVM network in artificial neural network. The identification effects of BP under the two types of characteristics are compared, and a large number of simulation experiments are preformed to compare the key parameters of different networks by using the LPC combination parameters, and determine the key parameter values. And in this thesis, the idea of cross validation is also introduced. In MATLAB 7.0 environment, the pattern recognition simulation experiments are achieved, and the results show the validity of the LPC combination parameters and the identification algorithm based on neural network.4. Build a simple GUI demo platform. A simple GUI demo platform is built to show the waveform features and the effect of the identification for different patterns.5. Design the online recognition system based on VC ++6.0. The system is implemented by setting the threshold and the difference in algorithm. The results show that the identification effect is obvious to particular persons, and the system can identify different saccade angles and make feedback.In a word,the saccadic angles based on EOG are identified and classified in this thesis. By designing experiments, recording the data, doing some pretreatments such as filtering and endpoint detection to the original EOG signals, and then using the effective parameters for pattern recognition, the good identification results are achieved. At last, the online system based on different saccade angles is designed and implemented, which is the basic work for further research of EOG-based HCI system.