Robust electrooculography endpoint detection based on autoregressive spectral entropy

Bio-based human computer interface has become a research hotspot in recent years. Accurate Electrooculography (EOG) endpoint detection is important for EOG pattern recognition. In the current paper, Autoregressive (AR) spectral entropy algorithm is proposed for EOG endpoint detection. Based on the analysis of EOG spectrum features, traditional Fast Fourier Transform (FFT)-based entropy exists during spectral leakage, influencing the spectrum probability distribution and further decreasing the entropy-domain signal to noise. To solve this problem, the AR spectrum is used to replace the FFT spectrum, thus keeping the detection algorithm robust. Furthermore, asymmetric thresholds are used for adaptive on-line detection in the entropy domain. Experimental results based on real-life EOG signals reveal that the proposed algorithm has higher robustness and better accuracy than traditional FFT spectral entropy in low SNR conditions.