Print Email Facebook Twitter Constructing a region of interest using map information for object tracking in autonomous vehicles Title Constructing a region of interest using map information for object tracking in autonomous vehicles Author Stakelbeek, R.A.R. Contributor Jonker, P.P. (mentor) Domhof, J.F.M. (mentor) Happee, R. (mentor) Krasnov, O.A. (mentor) Faculty Mechanical, Maritime and Materials Engineering Department Biomechanical Engineering Programme Bio-Mechanical Design, Automotive Human Factors (BMD-AUT) Date 2016-06-29 Abstract The Delft University of Technology intends to aid in the development of autonomous vehicles by building their own experiment platform in the DAVI project. To be able to make valid decisions, an autonomous vehicle needs to know its surroundings. Many sensors are used to detect any objects within range, but the sensor information has to be processed to form a map of the environment with all objects of importance on it; this process is called tracking. Current methods of tracking and associating objects are accurate enough to be safely used in an autonomous vehicle, but require significant computation power. Increasing the speed of the algorithms without losing performance is a real challenge. As with many methods, the joint probabilistic data association (JPDA) algorithm gets exponentially slower with more objects and more measurements. It is therefore proposed that a method that splits the survey area into smaller regions could lead to a faster algorithm. In this research, a region is dynamically constructed while the vehicle drives around using extended maps that contain all the information on the infrastructure (eHorizon). The region is created by merging polygons that are defined by the road shape. Objects within the region are tracked with a JPDA algorithm, while objects further away (outside the region) are tracked with a fast and simple nearest neighbor Kalman filter. Once an object gets inside the region, the track will automatically be handled by the JPDA algorithm. Experiments show that the region tracking algorithm performs faster than a JPDA algorithm without clustering and manages to keep track of objects in challenging environments. Compared to a clustering JPDA algorithm the processing times are slightly higher, but the region trackers shows more robustness in densely cluttered scenarios where large clusters mean more processing time for the clustering algorithm. Subject region of interestROItrackingeHorizonautonomousvehicleself-drivingautomotiveJPDAinfrastructuremap data To reference this document use: http://resolver.tudelft.nl/uuid:62a97ff4-282d-4300-91fc-d3fb66049b56 Embargo date 2018-06-29 Part of collection Student theses Document type master thesis Rights (c) 2016 Stakelbeek, R.A.R. Files PDF Stakelbeek_Thesis_digital.pdf 3.47 MB Close viewer /islandora/object/uuid:62a97ff4-282d-4300-91fc-d3fb66049b56/datastream/OBJ/view