Artificial intelligence system for driver distraction by stacked deep learning classification

Abstract

Increasing efforts in the transportation system have recently improved driver safety and reduced crash rates. Lack of attention and fatigue directly affect the driver's consciousness. Driver distraction is an essential driver-specific factor in the practical applicability of forward collision warning (FCW). However, there are still too many false alarms generated by the existing FCW system to be mitigated. This paper proposes facial detection to identify features and test anomalies' prediction against drivers using stacked convolutional neural network (CNN) layers. The proposed model used overlapping HAAR and stacked CNN features to identify classifications of eye areas, such as open or closed. In addition to the sliding query window's overall intensity information. The conventional HAAR function, which elevates the brightness of nearby regions, is still preferable. This method considers current intelligent transportation system-based solutions to minimize distraction effects by continuously comparing with flexible thresholds. The experimental results are analyzed from accurate driving datasets. At 456 iterations, the results acquired over 80% accuracy, while loss is near zero. The implication of driver's risk tolerance is further explored in this manner. Several risks are connected to driving any type of transportation system.