Detection of point-shaped targets


G. Farnebäck, H. Knutsson, G. Granlund
LiTH-ISY-R-1921
December, 1996

Download full paper

Abstract

This report documents work done at the request of the Swedish Defense Research Establishment. The studied problem is that of detecting point-shaped targets, i.e. targets whose only significant property is that of being very small, in a cluttered environment.

Three approaches to the problem have been considered. The first one, based on motion compensation, was rejected at an early stage due to expected problems with robustness and computational demands. The second method, based on background modeling with principal components, turned out successful and has been studied in depth, including discussion of various extensions and improvements of the presented algorithm. Finally, a Wiener filter approach has also turned out successful, including an approximation with separable filters.

The methods have been tested on sequences obtained by an IR sensor.

While both the two latter approaches work well on the test sequences, the Wiener filter is simpler and computationally less expensive than the background modeling. On the other hand, the background modeling is likely to have better possibilities for extensions and improvements.

Reference

Farnebäck G, Knutsson H, Granlund G. Detection of point-shaped targets. Report LiTH-ISY-R-1921, Computer Vision Laboratory, SE-581 83 Linköping, Sweden, 1996.

Bibtex entry

@TechReport{farnebackREPORT96,
  author         = {G. Farneb{\"a}ck and H. Knutsson and G. Granlund},         
  title          = {Detection of point-shaped targets},                        
  institution    = {Computer Vision Laboratory},                               
  year           = 1996,                                                       
  type           = {Report},                                                   
  number         = {LiTH-ISY-R-1921},                                          
  address        = {SE-581 83 Link{\"o}ping, Sweden},                          
  month          = {December}
}