Using long short term memory and convolutional neural networks for driver drowsiness detection.

摘要: Fatigue negatively affects the safety and performance of drivers on road. In fact, drowsiness fatigue are cause a substantial number motor vehicle accidents. Drowsiness among can be detected using variety modalities, including electroencephalogram (EEG), eye movement, driving dynamics. Among these EEG is highly accurate but very intrusive cumbersome. On other hand, dynamics easy to acquire accuracy not high. Eye movement based approach attractive in terms balance between two extremes. However, techniques normally require an tracking device which consists high speed camera with sophisticated algorithm extract related parameters such as blinking, closure, saccades, fixation etc. This makes detection difficult implement practical system, especially embedded platform. this paper, authors propose use images from directly without need for expensive eye-tracking system. Here, movements captured by Recurrent Neural Network (RNN) detect drowsiness. Long Short Term Memory (LSTM) class RNN has several advantages over vanilla RNNs. work array LSTM cells utilized model movements. Two types LSTMs were employed: 1-D (R-LSTM) used baseline convolutional (C-LSTM) facilitates 2-D directly. Patches size 48 × around each extracted 38 subjects, participating simulated experiment. The state vigilance subjects independently assessed power spectral analysis multichannel (EEG) signals, recorded simultaneously, binary labels alert drowsy (baseline) generated. Results show efficacy proposed R-LSTM resulted 82 % C-LSTM range 95%-97%. Comparison also provided recently published approach, showing technique outperform wide margin.