Anti-Vice Recommendation Systems Based on Face Images Using Machine Learning

Abstract

Face attribute analysis is a valuable aid in biometric-based human identification system. It is a challenging task to perform due to variations in illumination, occlusion, pose and other variables with respect to a face presentation. This paper proposes a model for determining whether a driver has consumed alcohol or drugs, or is under the control of some other vices based on the images of the face taken at a specific period of time. In order to fulfill this task images of the face and parts of the face are captured by a camera device. These various parts of the face mainly the cheek, neck, eyes, ear, etc., are used to detect alcohol consumption based on deviations of the face with respect to color, change in size of eyes and other parts of the face. A three layer network of neurons is used for experimentation. The neural network is trained using a back propagation algorithm and accuracy is checked for the model by applying the test data to the trained model. The expected results from the model involve a prediction stating the detection of a vice.