Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Detektering, målföljning och identifiering med robotsystem 90

Lindgren, Simon (2015)
Department of Automatic Control
Abstract
The objective was to clarify whether and how it is possible to detect, track and identify with an EO/IR (electro optical/infrared) sensor. Moreover, the sight of robotic system 90 was tested in order to clarify the possibilities of using it as a sensor. The sight has a field of view of 3x4° with the TV camera and 4x6° with the IRV camera. Therefore, the project focused on a moving sensor, in order to be able to scan a larger area.

By background subtraction it´s possible to detect changes, i.e., movements. The background must move with the sight and therefore requires measurements of the sight direction to be accurate. Detection was made possible by noise reduction of the measured values using a Kalman filter. The maximum scan rate of... (More)
The objective was to clarify whether and how it is possible to detect, track and identify with an EO/IR (electro optical/infrared) sensor. Moreover, the sight of robotic system 90 was tested in order to clarify the possibilities of using it as a sensor. The sight has a field of view of 3x4° with the TV camera and 4x6° with the IRV camera. Therefore, the project focused on a moving sensor, in order to be able to scan a larger area.

By background subtraction it´s possible to detect changes, i.e., movements. The background must move with the sight and therefore requires measurements of the sight direction to be accurate. Detection was made possible by noise reduction of the measured values using a Kalman filter. The maximum scan rate of the sight was 10.8°² /s. The sight can thus not be used as a sensor because the scanning speed is judged to be too low.
Kalman filters may also be used to filter the detections and for a short time predict the target path. With a scaling matrix, error in the velocity vector was reduced when the target was not manoeuvring and the target movement could be predicted over time. However, it led to delays when the target manoeuvred, resulting in poor accuracy of the target position.

The video color from the sight is in grayscale, which means that the color couldn´t be used to identify aircraft. The aircraft is dark with the TV camera and bright with the IRV camera. What was left then was to identify the type of aircraft by analysing the shape.

An existing method for identification by shape analysis is HOG (Histogram of Oriented Gradients). A large amount of training data was required for the support vector machine to handle the differences in shape depending on the sensor viewing angle, the aircraft´s rotation and the relatively small pictures. However, the reliability of the method was still not high enough.
Two custom methods were therefore tested in order to try to identify targets by shape analysis. In the first method a 3D model of an airplane was built. The identification would be done by rotating and projecting the model to a two-dimensional image which was compared with the image of the target. Finding the best three-dimensional rotation is a problem which can be solved only by testing all solutions. Therefore, it takes too long to identify with this method.

The second method aims to analyse the shape by rotating the two-dimensional image so that the nose is pointing along the x-axis of a two-dimensional coordinate system. Wings were then analysed by analysing the values on the y-coordinates relative to the x-coordinates. Maxima, minima and the gradients were analysed in order to place the wings, estimate their relative size to each other and the shape of the aircraft.

Method two may be used to increase the reliability of the identification of the aircraft. An autonomous combat system requires the identification to be very reliable, thus requiring the use of an optical zoom to enlarge the images. More parameters such as position, speed and size should also be analysed when identifying the target. (Less)
Please use this url to cite or link to this publication:
author
Lindgren, Simon
supervisor
organization
year
type
H3 - Professional qualifications (4 Years - )
subject
ISSN
0280-5316
other publication id
ISRN LUTFD2/TFRT--5973--SE
language
Swedish
id
7471114
date added to LUP
2015-07-03 09:27:31
date last changed
2015-07-03 09:27:31
@misc{7471114,
  abstract     = {{The objective was to clarify whether and how it is possible to detect, track and identify with an EO/IR (electro optical/infrared) sensor. Moreover, the sight of robotic system 90 was tested in order to clarify the possibilities of using it as a sensor. The sight has a field of view of 3x4° with the TV camera and 4x6° with the IRV camera. Therefore, the project focused on a moving sensor, in order to be able to scan a larger area.

 By background subtraction it´s possible to detect changes, i.e., movements. The background must move with the sight and therefore requires measurements of the sight direction to be accurate. Detection was made possible by noise reduction of the measured values using a Kalman filter. The maximum scan rate of the sight was 10.8°² /s. The sight can thus not be used as a sensor because the scanning speed is judged to be too low.
 Kalman filters may also be used to filter the detections and for a short time predict the target path. With a scaling matrix, error in the velocity vector was reduced when the target was not manoeuvring and the target movement could be predicted over time. However, it led to delays when the target manoeuvred, resulting in poor accuracy of the target position.

 The video color from the sight is in grayscale, which means that the color couldn´t be used to identify aircraft. The aircraft is dark with the TV camera and bright with the IRV camera. What was left then was to identify the type of aircraft by analysing the shape.

 An existing method for identification by shape analysis is HOG (Histogram of Oriented Gradients). A large amount of training data was required for the support vector machine to handle the differences in shape depending on the sensor viewing angle, the aircraft´s rotation and the relatively small pictures. However, the reliability of the method was still not high enough.
 Two custom methods were therefore tested in order to try to identify targets by shape analysis. In the first method a 3D model of an airplane was built. The identification would be done by rotating and projecting the model to a two-dimensional image which was compared with the image of the target. Finding the best three-dimensional rotation is a problem which can be solved only by testing all solutions. Therefore, it takes too long to identify with this method.

 The second method aims to analyse the shape by rotating the two-dimensional image so that the nose is pointing along the x-axis of a two-dimensional coordinate system. Wings were then analysed by analysing the values on the y-coordinates relative to the x-coordinates. Maxima, minima and the gradients were analysed in order to place the wings, estimate their relative size to each other and the shape of the aircraft.

 Method two may be used to increase the reliability of the identification of the aircraft. An autonomous combat system requires the identification to be very reliable, thus requiring the use of an optical zoom to enlarge the images. More parameters such as position, speed and size should also be analysed when identifying the target.}},
  author       = {{Lindgren, Simon}},
  issn         = {{0280-5316}},
  language     = {{swe}},
  note         = {{Student Paper}},
  title        = {{Detektering, målföljning och identifiering med robotsystem 90}},
  year         = {{2015}},
}