Skip to Main Content
 

Global Search Box

 
 
 

ETD Abstract Container

Abstract Header

Traffic Scene Perception using Multiple Sensors for Vehicular Safety Purposes

Hosseinyalamdary , Saivash, Hosseinyalamdary
http://rave.ohiolink.edu/etdc/view?acc_num=osu1462803166

Abstract Details

2016, Doctor of Philosophy, Ohio State University, Civil Engineering.
Autonomous driving is an emerging technology, preventing accidents on the road in future. It, however, faces many challenges because of various environmental conditions and limitations of sensors. In this dissertation, we study multiple sensor integration to overcome their limitations and reliably perform missions enabling autonomous driving. The laser scanner point cloud is a rich source of information, suffers from low resolution, especially for farther objects. We generalize 2D super-resolution approaches applied in image processing to the 3D point clouds. Two variants are developed for the 3D super-resolution: the dense point cloud is generated in such a way that it follows the geometry of the original point cloud; the brightness of the images are utilized to generate the dense point cloud. The results show our proposed approach successfully improves the density of the point cloud, preserves the edges and corners of objects, and provides more realistic dense point cloud of objects relative to the existing surface reconstruction approaches. The static and moving objects must be detected on the road, the moving objects must be tracked, and trajectory of the platform must be designed to avoid accidents. The densified point clouds are integrated with other sources of information, including the GPS/IMU navigation solution and GIS maps, to detect the objects on the road and track the moving ones. The results show static and moving objects are detected, the moving objects are accurately tracked, and their pose is estimated. In addition to obstacle avoidance, the autonomous vehicles must detect and obey the traffic lights and signs on the road. Due to the variations in the traffic lights, we propose Bayesian statistical approach to detect them. The spatio-temporal consistency constraint is applied to provide coherent traffic light detection in space and time. In addition, conic section geometry is utilized to estimate the position of the traffic lights with respect to camera mounted on the platform. The proposed traffic light detection approach is evaluated using Karlsruhe Institute of Technology (KITTI) and La Route Automatise (LARA) benchmarks. The results of the proposed traffic light detection approach are 98.7% precision rate and 94.7% recall rate in LARA benchmark, outperform the existing traffic light detection approaches tested in LARA benchmark. In conclusion, we integrate multiple sensors to overcome their shortages, such as low resolution of point clouds, and propose obstacle avoidance and traffic light detection approaches based on the integrated sensors. Our results outperform the earlier studies in traffic light detection and provide more realistic surfaces in 3D super-resolution. Further studies may modify the proposed traffic light detection to detect the traffic signs.
Alper Yilmaz (Advisor)
Charles Toth (Committee Member)
Ralph von Frese (Committee Member)
166 p.
Civil Engineering
Multiple sensor integration; moving object tracking; surface reconstruction; super-resolution; traffic light detection; conic section geometry; autonomous vehicle; spatio-temporal consistency constraint; non-holonomic constraint;

Recommended Citations

Citations

  • Hosseinyalamdary , Hosseinyalamdary, S. (2016). Traffic Scene Perception using Multiple Sensors for Vehicular Safety Purposes [Doctoral dissertation, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu1462803166

    APA Style (7th edition)

  • Hosseinyalamdary , Hosseinyalamdary, Saivash. Traffic Scene Perception using Multiple Sensors for Vehicular Safety Purposes. 2016. Ohio State University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=osu1462803166.

    MLA Style (8th edition)

  • Hosseinyalamdary , Hosseinyalamdary, Saivash. "Traffic Scene Perception using Multiple Sensors for Vehicular Safety Purposes." Doctoral dissertation, Ohio State University, 2016. http://rave.ohiolink.edu/etdc/view?acc_num=osu1462803166

    Chicago Manual of Style (17th edition)

osu1462803166
148
© 2016, some rights reserved.Creative Commons License Traffic Scene Perception using Multiple Sensors for Vehicular Safety Purposes by Saivash Hosseinyalamdary Hosseinyalamdary is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License. Based on a work at etd.ohiolink.edu.
This open access ETD is published by The Ohio State University and OhioLINK.