Facial Feature-Based Drowsiness Detection With Multi-Scale Convolutional Neural Network

Recently, the upsurge in accidents is caused due to driver drowsiness arises due to lack of sleep, fatigue and other health factors. The drowsiness leads to mortality, loss of properties and serious health conditions. Hence, it is necessary to prevent accidents caused by drowsiness in drivers. At present, the automated model is effective for detection drowsiness and recognition. In this research paper, developed a MCNN (Multi-Scale Convolutional Neural Network) framework for the classification of drowsiness. Initially, the YAWDD dataset and NTHU-DDD dataset is utilized for acquiring video sequences about driving. The acquired video is converted into frames for the keyframe extraction and selection. With the Dlib library face recognition is utilized for the localization of facial points in the extracted frames. The extracted image frames are pre-processed with the Cross Guided Bilateral Filtering followed by the feature extraction with hybrid dual-tree complex wavelet transforms with Walsh-Hadamard transform for the extraction of the feature vector in the image frame blocks. The feature points are optimized with the Flamingo search algorithm (FSA) integrated with the deep learning model Multiscale convolutional neural network (MCNN). With the proposed method, MCNN based FSA model feature extraction drowsy and non-drowsy are classified. The proposed simulation results illustrated the MCNN with FSA model attains an accuracy value of around 98.38% for YAWDD dataset and NTHU-DDD model exhibits an accuracy value of 98.26%. The performance of the proposed MCNN model exhibits approximately 6% higher accuracy than the conventional state-of-art methods.