Intelligent reconnaissance functions 2016-2018, final report


  • Fredrik Näsström
  • Fredrik Bissmarck
  • Viktor Deleskog
  • Fredrik Hemström
  • Max Holmberg
  • Jörgen Karlholm
  • Jonas Nordlöf
  • Jonas Nygårds
  • Karl-Göran Stenborg
  • Niclas Wadströmer

Publish date: 2018-12-21

Report number: FOI-R--4648--SE

Pages: 35

Written in: Swedish


  • target recognition
  • collaboration
  • fusion
  • sensor control
  • positioning
  • machine learning


This report presents the work carried out in the project Intelligent reconnaissance functions. In the project, research has been conducted with the aim to make future sensor systems more intelligent. By being more intelligent, future sensor systems provide sensor operators with better situational awareness, higher system confidence, reduced risk of human errors and low mental workload. The project has conducted research in the area of automatic target recognition, positioning without using satellite navigation, and sensor planning. Within target recognition, two automatic target recognition algorithms have been developed for automatic detection and recognition of different objects. The developed target recognition algorithms are good at recognizing both moving and stationary objects in complex environments. The algorithms are based on deep neural networks. In the project we have worked with civilian objects, but the algorithms can learn to recognize other objects such as combat vehicles, ships and helicopters. Within automatic positioning without using satellite navigation, two tests have been made to analyze the positioning performance. In the first test, an individual vehicle has traveled along a path, after which the total positioning error is measured. In the second test, a target has been detected with an IR sensor and is measured from two self-positioning vehicles. The position of the target, which in this case was a person, has been estimated in a common coordinate system for the vehicles. The first test showed that the positioning system has good performance in different terrains with a small uncertainty, and the second test has showed that the relative uncertainty becomes smaller for two positioning systems that cooperate than the uncertainty for each of the individual positioning systems. The measurement uncertainty of a triangulated target is mainly due to relative angular uncertainty, and it becomes smaller with cooperative positioning system so the measurement uncertainty for the target decreases. Within reconnaissance with collaborating sensors, the project has demonstrated how these can be used to increase reconnaissance coverage and improve target track quality. The results show that sensors that are allowed to be controlled automatically can increase the reconnaissance performance. By automatically controlling and processing data at a high scan rate a higher track quality is achieved. Automatic sensor planning therefore gives a higher reconnaissance capability, because when a threat has been detected by any sensor, the sensor which maximizes target precision is selected and other sensors can continue to scout. This contributes to a more resource efficient use of available sensors.