Description
TitleScaling data sharing among vehicles to tackle long tail traffic situations
Date Created2021
Other Date2021-05 (degree)
Extent1 online resource (xi, 108 pages)
DescriptionDeveloping robust driver assistance and automated driving technologies requires an understanding of not just common highway and city traffic situations but also the long tail of unusual or rare traffic events (e.g., objects on the roadway and pedestrian crossing highway, etc.). Due to the rareness, variety, and complexity of the long tail traffic conditions, it is widely recognized that ensuring dependability under these situations remains a key challenge for automated vehicles. Specifically, the challenges of tackling the long tail traffic situations are threefold. (i) Due to the rareness of the long tail traffic events, understanding these events will require accumulating data collection in the order of billions of miles. Since most existing efforts to collect driving data build on a small fleet of highly instrumented vehicles that are continuously operated with test drivers, it is challenging to acquire the road data in such a large scale. (ii) Among the large amount of data needed, it is hard to automatically identify the unusual situations which are challenging to address and could lead to potential accidents. Although there exists a large body of work on abnormal driving event detection, they focus on detecting specific, known situations but cannot detect previously unknown unusual road events that are missing in the current set of test cases for automated vehicles. (iii) The complexity of diverse traffic situations makes a single vehicle hard to sense its surrounding comprehensively. Although vehicle-to-vehicle (V2V) communications provide a channel for point cloud data sharing, it is challenging to align point clouds from two vehicles with state-of-the-art techniques due to localization errors, visual obstructions, and viewpoint differences.
To address such challenges, this thesis focuses on scaling data sharing among vehicles, which enables low-cost road data acquisition, unusual driving events identification and accurate vehicle to vehicle point cloud alignment. In particular, the proposed solution includes: (1) A light-weight sensing and driving data logging system that can derive internal driver inputs (i.e., steering wheel angles, driving speed and acceleration) and external perceptions of road environments (i.e., road conditions and front-view video) using a smartphone and an IMU mounted in a vehicle, which enables road data acquisition and sharing in heterogeneous driving scenarios crossing different vehicle models and drivers. (2) An automatic unusual driving events identification system, which can detect unusual situations through a two-pronged approach involving inertial monitoring of driver reactions and an autoencoder-based technique for detecting unusual video scenes. (3) A two-phase point cloud registration mechanism to fuse point clouds from two different vehicles, which focuses on key objects in the scene where the point clouds are most similar and infer the transformation from those. The systems are evaluated based on both experimental and simulation data, which shows accurate road data acquisition and unusual event identification, and tremendous point cloud fusion performance improvement when combining with state-of-art baselines.
NotePh.D.
NoteIncludes bibliographical references
Genretheses, External ETD doctoral
LanguageEnglish
CollectionSchool of Graduate Studies Electronic Theses and Dissertations
Organization NameRutgers, The State University of New Jersey
RightsThe author owns the copyright to this work.
Statistical Profile