|
Performance assessment and modeling of fused systems is an open problem. Main reason is the strong scene and task dependency of fusion algorithms. The idea followed here is to divide the task in detection and situational awareness ones and analyze them separately. An experimental common line of sight system consisting of a LWIR-bolometer and a low light CMOS camera was set-up to do so. Its first use was detection performance assessment. A data collection of a human at different distances and positions was analyzed in a perception experiment with the two original bands and eight different fusion methods. Twenty observers participated in the experiment. Their average detection probability clearly depends on the imagery. Although the resolution in LWIR is three times worse than the visual one, the achieved detection performance is much better. This transforms in the fused imagery also, with all fusion algorithms giving better performance than the visual one. However, all fusion algorithms to a different degree decrease observer performance compared to LWIR alone. This result is in good agreement with a Graph-Based Visual Saliency image analysis. Thus, it seems possible to assess fusion performance for the detection task by saliency calculations. |
ACCESS THE FULL ARTICLE
No SPIE Account? Create one
Night vision
Image fusion
Long wavelength infrared
Visualization
Situational awareness sensors
Bolometers
Goggles
