Low-Cost and Accurate 3D Road Modeling Using Mobile Phone

This paper studies the feasibility of using mobile phone to achieve low-cost and accurate three-dimensional (3D) road modeling, which provides nutritious supplement to current digital map. Instead of using expensive tools, our method only needs a mobile phone fixed on a moving vehicle. Road surface variation, reflected by vehicle dynamic state, is captured by sensors in the mobile phone.The road in 3D is reproduced by combining sensing data from accelerometer, magnetometer, gyroscope, and GPS trace. Several compensation and calibration techniques are developed to improve modeling accuracy based on the systematic study of sensor error properties. A learning step is used to automatically unify the user dependent factors to avoid manual adjustment. We evaluate our method on a 2 km real road. Compared with the field measured baseline, it successfully generates a fine-grained 3D road model with maximum height error less than 4 m and 90 degree slope angle error less than 2 degrees.