*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
Ph.D. Thesis Proposal by
(Advisor: Dr. Eric N. Johnson)
Friday, June 9, 2017 @ 11:00 a.m.
Montgomery Knight Building - Room 317
Abstract:
Obstacle avoidance is one of the most important and expensive topics in Autonomous Robotics because it is an essential task for an autonomous vehicle to find an obstacle and avoid it. Any robotics designers use LIDAR, stereo camera and so on to be able to obtain more accurate sensing data. However, these approaches are more expensive than a single camera and need a specific device. Currently, many fields use UAVs for many objectives. Especially, photographing uses it from movie filming to personal use, and they have a single camera. Therefore, although the monocular camera-based recognition technique is a challenging task, its technique can be applied to a wider range, like an existing vehicle and a cheaper vehicle with fewer sensors. For this reason, this proposal mentions the monocular camera-based obstacle avoidance technique.
In this work, this technique can be separated into two parts: a mapping and a path planning. The mapping technique is designed by a monocular vision as input, which consists of the filter technique to obtain a point cloud and an occupancy grid map. The filter includes the pre-filter for initial convergence and an Extended Kalman Filter (EKF) for final localization and mapping. The occupancy grid map updates a probability of obstacle existence, and the octree data structure adds scalability and efficient memory management to the occupancy grid map. The path planning algorithm is designed based on the octree data structure, which consists of three parts: a new A* algorithm, a tree-based trajectory generation technique, and an additional lateral trajectory. The first A* algorithm can work on the octree data structure and outputs a free corridor. The second technology generates a trajectory in that corridor. The additional lateral trajectory improves the obstacle detection.
Committee Members:
Dr. Eric N. Johnson
Dr. Eric Marie J Feron
Dr. Jonnalagadda V R Prasad
Dr. Hao-Min Zhou
Dr. Patricio Antonio Vela
