Dynamic object tracking and classification from a moving platform

This thesis presents an object-level SLAM system capable of tracking objects in frame and classifying stationary and moving objects. The system combines two open-source algorithms, Mask R-CNN and ORB-SLAM. Mask R-CNN provides instance-level object detection and segmentation, while ORB-SLAM provides keypoint detection, camera tracking, and local mapping. A typical SLAM system assumes a static environment and treats dynamic objects in the scene as outliers. By using object-level information from Mask R-CNN, we extend the capability to recognize and track dynamic objects. The system uses only monocular images as input, resulting in numerous potentially low-cost applications without the need for calibrating multiple sensors or using a stereo rig. This system gives a mobile agent the capability of understanding and potentially interacting with its dynamic environment.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2021

Author dc:creator

• Lai, Andy

Contributors dc:contributor

• Do, Minh N

• Computer vision
• object recognition
• object tracking
• SLAM

Statement dc:rights

• Copyright 2020 Andy Lai

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/109359

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/109359