ENVIRONMENT MAPPING FOR AUTONOMOUS VEHICLES USING PARTICLE FILTER BASED ON ROAD BOUNDARY DETECTION

• Main Author: Adib Rasyidy, Mukhlas
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/79997
• Institution: Institut Teknologi Bandung

This research presents an alternative map representation for autonomous vehicles environment mapping that is based on particle representation and estimated with particle filter. In contrast to the well-developed grid-based map (known as OGM), the proposed particle map is a constructed sparsely and therefore requires less information that has to be maintained during mapping. The information contained in the proposed map is also limited to only road boundaries information, making the system does not have to memorize unnecessary data in the map. Therefore, this new kind of map is named as road boundary map (RBM) To create the road boundary map, the system must be capable of detecting the road boundaries from sensors’ data. The designed road boundary detection system uses camera and LiDAR data to obtain road boundary points in world coordinate system. One important element of that system is a new technique to project road pixels from camera to the world coordinate system, namely fitted plane projection (FPP). By having the road boundary detection results from camera and LiDAR in the same coordinate system, a data fusion for mapping purpose can be performed properly on both sensor’s data. The system was tested in Carla Simulator, a simulation software dedicated for autonomous vehicles research. The ground truth data for quantitative analysis are extracted by using a series of image processing techniques that is applied to data from Carla Simulator. This method allows us to extract road boundaries ground truth data with high resolution accurately and automatically. The evaluation results show that the proposed FPP was performing better than the existing transformation techniques. With this technique, road boundary detection system is able to detect the road boundary points with average error 0.17 m for LiDAR-assisted camera-based detection and 0.27 m for LiDAR-based one. Furthermore, the presented particle-based mapping system, RBM, can reduce the error further 0.20 m in average while increasing the precision of the local OGM.