In this paper we present the results of infrared processing and sensor fusion obtained within the European research project GEODE (Ground Explosive Ordnance DEtection) that strives for the realization of a vehicle-mounted, multi-sensor anti-personnel land-mine detection system for humanitarian demining. The system has three sensor types: an infrared camera, a ground penetrating radar and a metal detector. The output of the sensors is processed to produce confident values on a virtual grid covering the test bed. A confidence value expresses a confidence or belief in a mine detection on a certain position. The grid with confídence values is the input for the decision-Ievel sensor fusion and provides a co-registraiion of the sensors. We describe the methods TNO-FEL developed for the processing of infrared (3-5 µm) data to produce confidence values. We show results of experiments with infrared processing and sensor fusion on real time data. The performance of the processing and fusion are measured with the SCOOP evaluntion method that yields a less biased probability of false alarm by taking into account the spatial arrangement of false alarms.