Pothole Detection System using Machine Learning on Android

This paper investigates an application of mobile sensing: detection of potholes on roads. We describe a system and an associated algorithm to monitor the pothole conditions on the road. This system, that we call the Pothole Detection System, uses Accelerometer Sensor of Android smartphone for detection of potholes and GPS for plotting the location of potholes on Google Maps. Using a simple machine-learning approach, we show that we are able to identify the potholes from accelerometer data. The pothole detection algorithm detects the potholes in real-time. A runtime graph has been shown with the help of a charting software library ‘AChartEngine’. Accelerometer data and pothole data can be mailed to any email address in the form of a ‘.csv’ file. While designing the pothole detection algorithm we have assumed some threshold values on x-axis and z-axis. These threshold values are justified using a neural network technique which confirms an accuracy of 90%-95%. The neural network has been implemented using a machine learning framework available for Android called ‘Encog’. We evaluate our system on the outputs obtained using two, three and four wheelers. Keywords— Machine Learning, Context, Android, Neural Networks, Pothole, Sensor