Imaging Base Localisation for Dismounted Persons

This project will tackle issues in terms of accurate vision based self-localisation and relative motion navigation for dismounted soldiers GPS denied military missions. The solution includes mainly a stereovision setup of different modalities, visual and infrared, and other hardware capabilities conforming to the mission constraints. These will be required to run imaging algorithms processing the acquired data and outputting motion parameters required for self-localisation. The project is a combination of software development and hardware integration where software will include innovations in terms of image processing techniques across different modalities and providing precise motion estimation using a robust optimisation framework. Soldiers could be equipped with a visible camera and another infrared camera for night vision. The concept we propose here exploits the availability of the second infrared camera to complement the information obtained through the visible camera, in a stereo setup, to deliver a GPS free navigation solution avoiding constraints that we could find in monocular egomotion based solution. The challenge then is how to effectively use and process different imaging modality data and unifying their respective representations to contribute as feeds for the localisation step. The concept we propose will be able to extend this navigation solution into situations where night localisation is required. It opens the doors for solutions working under difficult constraints like navigating in complex night-time and also urban environment where the availability of ready map to be used is not possible. the system would be designed and miniaturised to fit the constraints of payload and space availability of dismounted soldiers type of application.
The ability to localise persons within the World is currently limited to the use of GPS or other active transmission based technologies. For the most part these technologies have a low accuracy and coverage is severely limited when inside buildings. Inertial sensor technologies can be used within very limited operational scenarios however these technologies are currently very susceptible to drift and therefore cannot be relied upon in isolation.
Localisation by analysing real-time video data from sensors mounted on the person shows significant promise for dead-reckoning localisation. Video imagery is not susceptible to drift or distortion from magnetic sources. Once the system has been calibrated from a known source such as GPS then localisation should be relatively error free. The challenge is to understand the distance to objects in the imagery, detect the relative movement of objects in the imagery and to capture imagery in extreme lighting conditions.
The project will research the use of multiple man mounted video sensors covering a range of optical bandwidths to undertake dismounted person localisation. This will include research into algorithms and software that can process the real-time imagery and provide accurate, low latency, reliable localisation data. Operational envelopes that will be considered will include day and night lighting conditions as well as scenes with close range objects (such as a built up area or inside a building) and long range objects (such as in a field or desert). The challenge will be to develop a non-intrusive and light weight system.