Precise Dead-reckoning Based on Multi-sensor Fusion

In this study, a method of accurate dead reckoning in urban environments via integration of a mono-camera, an IMU (gyro, wheel odometer), and GPS is proposed. Recently, robust and accurate vehicle position information has become essential for driver-assistance systems. In particular, dead reckoning plays an important role for the continuous positioning in areas, such as urban canyons, in which GPS positioning is limited. Visual odometry (VO) using a mono-camera is a method of dead reckoning which offers the potential of low cost and high performance, but has an accumulated heading error along the distance traveled. This study presents a method of heading estimation using GPS Doppler which corrects the accumulated heading error of VO. The key point in this study is the time-series tightly coupled integration of GPS Doppler and IMU that contributes to the improvement of availability of heading correction in an area of poor satellite reception, where heading estimation is not available by GPS Doppler only. As a result of evaluation in an urban environment, the error of dead reckoning is reduced to approximately 1/4, as compared to the conventional method, owing to the improvement and availability of heading correction in the urban environment.