Dissertation Announcement for Suvash Sharma – 03/04/2022 at 3:00 PM
February 23, 2022
Dissertation title: Performance enhancement of wide-range perception issues for autonomous vehicles
When: Friday, March 4, 2022, 3 PM – 5 PM
Where: https://msstate.webex.com/meet/ss3795
Candidate: Suvash Sharma
Degree: Ph.D., Electrical and Computer Engineering
Committee Members:
Dr. Bo Tang (Major Professor)
Assistant Professor, Department of Electrical and Computer Engineering
Dr. John Ball (Co-Major Professor)
Associate Professor, Department of Electrical and Computer Engineering
Dr. Yaroslav Koshka (Committee Member)
Professor, Department of Electrical and Computer Engineering
Dr. Junming Diao (Committee Member)
Assistant Professor, Department of Electrical and Computer Engineering
Abstract:
In this dissertation, we propose several novel algorithms dedicated for increasing the perception capability of autonomous vehicles. Specifically, scene understanding algorithms in-terms-of semantic segmentation are devised. While doing so, we present the effect of adverse weather condition that is commonly encountered in the driving world and propose the methods to minimize its’ harmful effects to the segmentation performance. More specifically, first, we propose a transfer learning technique in order to transfer the knowledge from data-rich domain to data-scarce off-road driving domain for semantic segmentation such that the learned information from one domain could be efficiently transferred to another domain while reducing run-time and increasing the accuracy. Second, the performance of several segmentation algorithms is assessed under the easy-to-severe rainy condition and two methods of enhancing the robustness are proposed such that the scene understanding performance is minimally affected. Third, rather than using the rainy images directly with the perception algorithms, we propose a new method of eradicating the rain effect from the input images. Since autonomous vehicles are rich in sensors and each of them has the capability of representing different type of information , it is worthy to use the information from all the possible sensors and fuse it so that the environmental knowledge is better represented. Forth, we execute such fusion mechanism with a novel algorithm that facilitates the use of local and non-local attention in a cross-modal scenario with RGB camera images and lidar-based images for the road detection using semantic segmentation. Fifth, a conceptually new method of off-road driving trail representation is introduced which is called as Traversability. Under this concept, the definition of safer travel for a vehicle along an off-road trail is different since the same trail area which is easier to traverse for one vehicle type could be harder for another type. To establish this correlation, we introduce a new dataset called CaT (CAVS Traversability). This dataset is very helpful for future research in several applications including military purposes, robotic navigation, etc.