Adaption of Robotic Approaches for Vehicle Localization

Localization of a robot is a central theme of different elaborations. But the determination of an exact position for a vehicle in urban area is more complex and an extensive algorithm challenge. This paper adresses the precise localization of a vehicle by combination of different sensor information using a Kalman and Particle filter. We present a vision based approach, which handles especially the kidnapping problem by standard sensor technology of a series-production vehicle. Therefore, the fusion of GPS 1 data and odometry, like speed and turn rate, produces a good initial position, which is the basis for generating different pose hypotheses. The inaccuracy of the GPS receiver limits their range of dispersion. For the evaluation, the subsequent Particle filter matches high precise map information with the current grayscale image of a built in camera. Different approaches known from robotics engineering are validated for their feasibility in automotive applications.