A winter storm warning remains in effect for Hartford County until 7 p.m. today. Residential students are asked to postpone their return to campus to Monday afternoon or, for those whose class schedules allow, Tuesday. Students, please do not travel to and arrive on campus today.
Hartt Community Division activities today and this evening are canceled.
Thesis Title: Benchmarking of Visual Tracking Algorithms during Rapid UAV Motion
Student: Eric Jacobson
Degree: Master of Science
Dept.: Mechanical Engineering
Research Group: Autonomous Mobile Robotics Research Group
Date: Tuesday, December 18th
Time: 2:30 to 4pm
Location: UT 215
Dr. Akin Tatoglu, ME Department
Dr. Eoin King, ME Department
Dr. Kiwon Sohn, ECE Department
Autonomous mobile robots including unmanned aerial vehicles (UAV) fuse camera and inertial measurement unit data for perception, localization and mapping purposes. Localization algorithms require to identify a visual feature and to register the same feature in the next camera frame. The performance of landmark tracking depends on the processing time and total displacement of the robot between frames. If the robot has an agile motion plan –such as an emergency response robot or a co-robot at a human occupied environment— the physical displacement will be greater which will minimize the allowed processing time due to increased search area.
There are distinct problems when implementing such tracking algorithms. During rapid motion, the variables including viewing angle, orientation, distance and ambient lighting conditions will vary substantially. The goal of this thesis is to find out how well the gradient based and more advanced landmark detection and tracking algorithms work under highly dynamic conditions. This means to figure out what will happen if the UAV/camera is moved rapidly in an unexplored environment without any known landmarks or a priori map.
In order to simulate rapid drone motion, a robotic arm assembly and a simulated outdoor environment was created and used as the experimental setup to collect images. Mathematical background, robotic arm control logic, sensor fusion techniques, the experimental results of ten different motion profiles and ambient lighting conditions are plotted and compared with total 201 plots. It is shown that based on the path planning algorithm, a UAV can optimize and switch between algorithms with respect to the controller input.