19-04-2014, 04:56 PM
Object Detection and Tracking in Images and Point Clouds
Object Detection.pdf (Size: 1.58 MB / Downloads: 520)
ABSTRACT
This project was an attempt at developing an object detection and tracking system using
modern computer vision technology. The project delivers an implemented tracking system.
It consists of a hybrid of optical and modern infra-red technology and is applicable to areas
such as unsupervised surveillance or semi-autonomous control. It is stable and is applicable
as a stand alone system or one that could easily be embedded into an even larger system.
The project was implemented in 5 months, and involved research into the area of computer
vision and robotic automation. It also involved the inclusion of cutting-edge technology of
both the hardware and software kind. The results of the project are expressed in this report,
and amount to the application of computer vision techniques in tracking animate objects in
both a 2 dimensional and 3 dimensional scene.
Preamble
This project involved implementing real-time tracking software. This chapter provides a
discussion of the project specification. It also gives a high-level overview of the system,
leaving design and implementation details for discussion in the respective chapters. It also
provides a “roadmap” for the reader about the overall presentation and structure of the
report.
Initial Specification & Overview
The project goal was to produce a working system for tracking objects in 3 dimensional space.
The aim of the project was to begin from this spec, and design a solution to the problem. After
a satisfactory solution was designed, the task came to implement the solution. Throughout
the project, many problems arose. These problems varied from performance issues related to
code, and from implementation issues related to limitations of software technologies used. All
attempts at overcoming these problems are discussed in this report. Also, as is common case
with research, as the project progressed, the resulting tracking system presented in this report
differs from the initial design specification; it is more generic and focuses on not tracking pre-
empted objects but any objects that enter the sensor field of view (fov). Chapters 3 and 4
describe the design and implementation of the project and discuss the approach taken to go
from specification to the system presented alongside this report.
Tracking
As stated, the goal of this project was creating tracking software. This task is an ongoing topic
for research, with continous development of algorithms and implementation code. Results
of this research have been applied to a range of products such as video surveillance, robotic
vision and autonomous flight. In robotics, tracking is frequently used to provide a means
for localization and mapping of an unknown environment. One example of this has been the
application to enable robots to act human-like and play games like soccer [DOL+ 05]. The
tracking system implemented with this project incorporates the use of the Kinect sensor to
provide real-time depth analysis. The sensor is described in more detail in 2.2.1.
Algorithms and Research
As well as understanding the software and hardware technology used in computer vision,
extensive background reading was done on the algorithms and techniques widely used in
computer vision applications. This section gives an overview of some of the papers read and
provides citations of the papers noted. For some sections that require some extra explaining,
the reader is referred to the appropriate appendix section.