Smart AVDNet: alcohol detection using vehicle driver face

Driver drowsiness and alcohol consumption leads for the majority of road accidents. Intoxicated or drugged drivers are less likely to make timely judgments, react quickly, and operate their vehicles safely. It also affect vision, coordination, and balance, making it difficult to maintain control of the vehicle. To address these primary causes, a real-time, cost-effective, and non-intrusive technology has been developed using powerful deep learning technology. In this research paper a novel Smart AVDNet has been proposed for alcohol detection in the face of car drivers using IMDB-WIKI dataset. The input images are taken from the dashboard camera, the captured images are pre-processed using adaptive unsharp mask guided filter. In the first level of checking, fuzzy based HOG-SSD approach is utilized whether the driver drunk based on the analysis of the eye and mouth. After analysis, drivers face based on the fuzzy decision is send as input to the second level. In this level ROI based CNN techniques by analysing the chin and cheek features of the car drivers. If the driver drunk the car stops automatically and whether the driver is not drunk the camera is keep monitoring. The Smart AVDNet demonstrated an average classification accuracy of 99.39%. When compared with Two-Stage DNN, CNN, and DNN, Smart AVDNet achieved accuracies of 5.39, 17.73, and 0.42%, respectively.