Aplicación para teléfono móvil con sistema operativo Android que permita detectar somnolencia y emitir una alarma sonora a conductores de automóvil mediante procesamiento de imágenes

Abstract

This document shows the methodology and implementation of a solution to a specific problem that is to give early warning to drivers who are in a state of sleepiness, this in order to avoid or reduce road accidents caused by this cause. A mobile phone application was developed that, through digital image processing, will allow an estimate of the state of sleepiness of the driver through the analysis of the eyes, and depending on the latter, give the user an audible alert. The mobile application was developed in the Android Studio development environment, and Opencv and Dlib image processing libraries were implemented. For the development of the mobile application in the first instance, a theoretical contextualization of the types of algorithms implemented for the detection of drowsiness was carried out, with this a preselection of the most convenient to implement was made, subsequently a subdivision of the total system to be implemented was carried out, in where they determined 4 critical sequential stages present which are: user face detection, eye detection and sleepiness detection. Following the sequence of the 4 stages described, the Viola Jones algorithm was selected as methods to perform the face detection, and the HOG (Histograms of oriented gradients), for the detection of the eyes the viola jones algorithm was selected, and the detection of landmarks, subsequently to detect flickering the pixel density method, Viola Jones was implemented due to lack of detection and the EAR index, and finally to detect the presence of drowsiness PERCLOS (percentage of closed eyes) was used. Based on the algorithms selected for each of the 4 stages mentioned above, 3 mobile applications were made that were structured as follows: The first application used the VJ algorithm for face detection, eye detection was also performed using VJ and for the detection of flickering the pixel density analysis was implemented. For the second, the Viola Jones algorithm (VJ) was implemented as a method for face detection, to detect the Viola Jones algorithm again and for the detection of flickers, a Viola Jones algorithm characterized by only detecting was used Open eyes. Finally, for the third application, face detection was performed using HOG, eyes were detected using landmarks and the EAR index was implemented for the detection of flickering. Finally, for the three applications implemented, it was used as a method of detecting sleepiness PERCLOS Having all three applications developed, tests were carried out with parameters such as processing times, stability in uncontrolled light environments and number of successes and errors in the detection of drowsiness. Subsequently, the most efficient methods were selected and based on them the final application was developed, which ended up structured as follows: Face detection was performed using the VJ algorithm, eye detection was performed using Landmarks, they were detected the blinking by means of EAR and the state of sleepiness was estimated by PERCLOS. A sound alert was implemented that will be activated when the PERCLOS sleepiness index is greater than 80%. Functional tests of the application were carried out in the application environment (vehicle), where the correct location of the mobile device was estimated for optimum operation of the application, as well as the range of lighting where it works optimally. Subsequently, the operation of the application was estimated according to user characteristics such as gender, presence of facial features such as the presence of a beard, and the use of accessories such as glasses. Limits were found in the correct operation of the application in users with glasses. The mobile application was designed for cell orientation in Portrait mode (closeup), with an image resolution of 640x480 pixels, the minimum Android version required for the operation of the application is Android 4.0. In the graphical user interface you can display parameters such as face detection, the PERCLOS index, and two buttons that allow you to perform the application calibration for optimum operation.