Lane Marking-Based Vehicle Localization Using Low-Cost GPS and Open Source Map
Résumé
Autonomous Vehicles applications and Advanced Driving Assistance Systems (ADAS) need scene understanding processes allowing high level systems to carry out decision. For such systems, the localization of a vehicle evolving in a structured dynamic environment constitutes a complex problem of crucial importance. However, the low accuracy of the GPS system in urban environments makes its localization unreliable for further treatments. The combination of GPS data and additional sensors (WSS, IMU or Camera) can improve the localization precision. More and more, digital maps are also used in this process. Generally, these maps are customized or built for a specific application, asking high cost to design and upgrade. In this article, we propose a low cost localization system based on camera, GPS and open map. Starting from the road marking, detected by a multi-kernel estimation method, a particle filter generates the samples taking advantage of lane markings to predict the most probable trajectory of the vehicle and the low cost GPS position. Then, the accuracy of the localization is improved using an open map. This process was validated through several scenarios with a public database and our experimental platform.