Real-time aerial vehicle detection and tracking with depth-aided vision sensing

We study the problem of detecting and tracking flying objects in real-time with color and depth images. We improve the sparse part-based representation learning approach by utilizing depth data from depth vision sensor to achieve much faster detection speed while maintain high detection accuracy. We revised some of algorithms presented in part-based representation method to get marginally better performance. Then we invented a novel data preprocessing method, which is based on edge detection and contour selection to generate possible vehicle locations before the image is processed by classifier. This approach can be applied to any object with distinguishable parts in relatively fixed spatial configurations, and our target here is the flying vehicle at indoor environment. Since flying objects tend to change poses and locations fast and frequently, the detection algorithm needs to run fast so that the tracking algorithm can keep on tracking the detected object. We also use hardware acceleration tools to further increase algorithm speed. The results of vehicle localization and tracking are shown and a critical evaluation of our approaches is also presented.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Mechanical Engineering

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2015

Author dc:creator

• Zhang, Bicheng

Contributors dc:contributor

• Dullerud, Geir E.

• real-time
• aerial vehicles
• drone
• depth sensor
• Kinect
• object recognition
• tracking
• detection

Statement dc:rights

• Copyright 2015 Bicheng Zhang

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/88127

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/88127