Advanced

Car detection with and without motion features

Sundbeck, Kajsa LU (2016) In Master’s Theses in Mathematical Sciences FMA820 20162
Mathematics (Faculty of Engineering)
Abstract
Object detection is a big research area and has been investigated for many different purposes and objects, such as faces, pedestrians and vehicles. Depending on the application there are different limitations to adjust to, but also possibilities to take advantage of.

The purpose of this thesis is to investigate the improvement of an existing car detector when the detections are performed in video sequences. The original detector uses only individual frames and the new one utilizes the video format by adding motion features in the detection.

Motion features are features which give information about motion in the image. In this work the motion features used come from background images which are calculated from previous frames in the... (More)
Object detection is a big research area and has been investigated for many different purposes and objects, such as faces, pedestrians and vehicles. Depending on the application there are different limitations to adjust to, but also possibilities to take advantage of.

The purpose of this thesis is to investigate the improvement of an existing car detector when the detections are performed in video sequences. The original detector uses only individual frames and the new one utilizes the video format by adding motion features in the detection.

Motion features are features which give information about motion in the image. In this work the motion features used come from background images which are calculated from previous frames in the video. The input to the detector is both the image and the background image and there is movement in the image where the images differ.

When evaluation was performed on the training data the detectors performed better with than without motion features. That is an indication that there might be additional information in the motion features.

On the validation data it was observed that when using motion features only moving objects were detected but although the evaluation was only performed in an area of the image where the vehicles were always moving there was no significant improvement to be found from the use of motion features. (Less)
Popular Abstract (Swedish)
För att automatiskt kunna utvärdera säkerheten i trafiken vill vi kunna hitta hur trafikanterna förflyttar sig i förhållande till varandra. Det första steget för att åstadkomma detta är att hitta — eller detektera — trafikanter i de enskilda bilderna i en film. Mycket har forskats på detektionsproblem i enstaka bilder men eftersom tillämpningen kommer att vara på filmer har det här tagits till vara på den information som finns i videoformatet genom att använda att det vi letar efter rör på sig.
Please use this url to cite or link to this publication:
author
Sundbeck, Kajsa LU
supervisor
organization
course
FMA820 20162
year
type
H2 - Master's Degree (Two Years)
subject
keywords
image analysis, object detection, machine learning, motion features, car detection
publication/series
Master’s Theses in Mathematical Sciences
report number
LUTFMA-3308-2016
ISSN
1404-6342
other publication id
2016:E52
language
English
id
8896216
date added to LUP
2017-01-30 16:01:51
date last changed
2017-01-30 16:01:51
@misc{8896216,
  abstract     = {Object detection is a big research area and has been investigated for many different purposes and objects, such as faces, pedestrians and vehicles. Depending on the application there are different limitations to adjust to, but also possibilities to take advantage of.

The purpose of this thesis is to investigate the improvement of an existing car detector when the detections are performed in video sequences. The original detector uses only individual frames and the new one utilizes the video format by adding motion features in the detection.

Motion features are features which give information about motion in the image. In this work the motion features used come from background images which are calculated from previous frames in the video. The input to the detector is both the image and the background image and there is movement in the image where the images differ.

When evaluation was performed on the training data the detectors performed better with than without motion features. That is an indication that there might be additional information in the motion features.

On the validation data it was observed that when using motion features only moving objects were detected but although the evaluation was only performed in an area of the image where the vehicles were always moving there was no significant improvement to be found from the use of motion features.},
  author       = {Sundbeck, Kajsa},
  issn         = {1404-6342},
  keyword      = {image analysis,object detection,machine learning,motion features,car detection},
  language     = {eng},
  note         = {Student Paper},
  series       = {Master’s Theses in Mathematical Sciences},
  title        = {Car detection with and without motion features},
  year         = {2016},
}