Strabismus free gaze detection system for driver’s using deep learning technique

Negligence driving behavior is one of the principal reasons of vehicle collisions as well as fatal deaths on road. The proper clues about gaze zone can play a significant role to justify the driver’s state of alertness. Although, supervising real-time driving performance exhibit some challenges such as distraction after long journey, lighting variations, eye recognition under sun glasses and so on. Including these challenges, there is a trend to issue the fake driving licenses by road authorities, even though the driver has already been suffering from vision problem as well as problem of occluded eye due to maximum pose variation. To resolve these issues, our proposed work scrutinize such frames which persist vision trouble primarily in the form of strabismus as well occluded eye frame, and filtered those frames from the dataset. Further, Histogram equalization is applied over each region of interest (ROI) image to diminish the light effect. Thus, strabismus containing frame as well as occluded eye frame removal and normalization via Histogram equalization upgrade the overall functioning of our model. Afterward, we have employed three parallel convolution neural networks (CNN) to capture features from such an absolutely refined ROI frames and Euclidean distances were computed based on the these features. Further, fusion scores were calculated over these estimated distances to classify the driver’s gaze zones. The proposed model is justified on open Columbia gaze dataset CAVE-DB. The designed method revealed the better accuracy in contrast of the earlier gaze categorization method.