Date of Award
5-2023
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Electrical and Computer Engineering
Committee Chair/Advisor
Melissa Smith
Committee Member
Yongkai Wu
Committee Member
Jon C. Calhoun
Abstract
Autonomous vehicle (AV) development has become one of the largest research challenges in businesses and research institutions. While much research has been done, autonomous driving still requires extensive amounts of research due to its immense, multi-factorial difficulty. Autonomous vehicles rely on many complex systems to function, make accurate decisions, and, above all, provide maximum safety. One of the most crucial components of autonomous driving is the perception system.
The perception system allows the vehicle to identify its surroundings and make accurate, but safe, decisions through the use of computer vision techniques like object detection, image segmentation, and path planning. Due to the recent advances in deep learning, state-of-the-art computer vision algorithms have made exceptional progress. However, for production-ready autonomous driving, these algorithms must be nearly perfect. Furthermore, even though perception systems are a widely researched area, most research focuses on urban environments and there exists a great need for autonomy in other areas. Specifically, autonomy for unmanned ground vehicles (UGV) needs to be explored. Autonomous UGVs allow for a wide range of applications like military usage, extreme climate exploration, and rescue missions.
This research aims to investigate bottlenecks within a perception system of autonomous UGVs and methods of improving them. The primary investigation focuses on the inference speed of semantic segmentation using deep learning techniques. Unlike object detection, semantic segmentation provides a much better understanding of the environment by providing pixel-wise classification rather than only creating bounding boxes. However, semantic segmentation comes at a much higher computational cost. Secondly, this thesis looks at increasing the image transfer time from a mounted camera to a video processing unit (which serves the deep learning model) using lossy compression. Due to the nature of lossy compression, we must also understand how lossy compression affects the classification accuracy of semantic segmentation. Finally, the challenges faced and preliminary results from future work relating to temporal consistency are discussed.
Recommended Citation
Selee, Bradley, "Improving Inference Speed of Perception Systems in Autonomous Unmanned Ground Vehicles" (2023). All Theses. 4053.
https://open.clemson.edu/all_theses/4053