Crash Free- A Smart System to Detect, Prevent Drowsiness and Locate Accidents

Abstract

Due to increased number of vehicles in the present era, there are significant number of problems faced by the drivers. Drowsiness is a major problem among them. This causes many accidents, and those accidents are very dangerous which will affect the economic development of a country. There is a necessity to keep drivers safe from drowsiness while they are driving. Also, it is recommended to keep a tracking mechanism of any driver to locate them easily if they met with an accident. Thus, the proposed system CrashFree focus on both drowsiness prediction and accident detection. This project uses behavioral based drowsiness prediction techniques to identify the state of a driver. Numerous approaches have been used under this technique previously. But this study uses a technique where it has three prediction models which will predict the state of face, eye, and mouth separately. The prediction models are trained with public and private datasets using Convolutional Neural Network (CNN) combining Support Vector Machines (SVM). Initially the video feed of a driver will be captured by a camera and regions of face, eye, and mouth will be extracted separately. Then it will be passed to their own prediction models and each state will be obtained. The final decision whether a driver is drowsy or not will be occupied by combining each separate states. The system will give a continuous alert via mobile application once a drowsy state is identified. On the other hand, the system uses a two-sensor based Internet of Things (IoT) platform to suspect an accident on a vehicle. The vibration sensor and accelerometer sensor will be used along with Arduino UNO. The sensor values will be transferred to the server on real-time. The values will be compared, and an accident suspect prompt alert will be sent to the driver if an accident is suspected. If the driver failed to acknowledge to the alert an automatic notification will be sent to the driver’s friends circle, family circles, etc. The drowsiness prediction module was tested with some public and private datasets, and it work well with numerous conditions. Prediction models of eye state, mouth state and face state gave an average accuracy of 96%, 97% and 85% respectively. The overall experiment of the module gave an average accuracy of 93% where the drowsiness prediction lacks in low light environment. The accident detection module cannot be evaluated in the real-world. Therefore, the prototype has been tested based on some conditions. The proposed system CrashFree will make an impact on the respective research area, and it provides a solution to the above-mentioned issues.