Driver Helper: A Mobile Based Application to Predict Driver Behaviour Using Classification Techniques
Self-driving cars can reduce the number of accidents in comparison with the cars driven by a driver. According to a research 80% of automobile accidents occurs due to human error, which includes looking but not seeing, texting, internal distraction as well as external distraction. Every year approximately 1.3 million people are killed by road accidents. Monitoring the behaviour of driver and predicting the chances of occurrences of any mishap remains a challenge. Sensor based tracking and face-based tracking system using cameras can be followed in order to detect driver distraction in an efficient manner. When a driver can adopt to drive according to the vehicle performance, road structure and varying climatic situations, his driving will be safe. When he fails to adopt to the environment, the level of risk increases. It is required to monitor and predict the behaviour of driver. To overcome all these issues, we have developed an application named DRIVERHELPER which can virtually assist the driver by corelating the camera data with the sensor data. 4 sensors such as GPS, accelerometer, gyroscope and magnetometer were used for gathering real-time data. Data related to Harsh Acceleration, Harsh Braking, Out of Max speed, Harsh turn, Harsh Left turn and Harsh Right turn were studied. A classification task is performed with K-Nearest Neighbour (KNN), Support Vector Machine (SVM) and Decision Tree for filtering out the noise and taking the relevant data for prediction. It is found that decision tree acts as a best classifier and a mobile alert is send to the driver to bring back his attention..