Abstract
There is a paucity of published systematic research investigating object detection within the military context. Here, we establish baseline human detection performance for five standard military issued camouflage patterns. Stimuli were drawn from a database of 1242 calibrated images of a mixed deciduous woodland environment in Bristol, UK. Images within this database were taken during daylight hours, in summer and contained a PASGT helmet, systematically positioned within each scene. Subjects (20) discriminated between the two image types in a temporal 2AFC task (500ms presentation for each interval), with the detection scenario being the percentage of instances participants correctly detected the target. Cueing (cued/not-cued to target location), colour (colour/greyscale) and distance from the observer (3.5/5/7.5m) were manipulated, as was helmet camouflage pattern. A Generalized Linear Mixed Model revealed significant interactions between all variables on participant performance, with greater accuracy when stimuli were in colour, and the target location was cued. There was also a clear ranking of patterns in terms of effectiveness of camouflage. We also compare the results with a computational model based on low-level vision, and eye tracking data, with encouraging results. Our methodology provides a controlled means of assessing any camouflage in any environment, and the potential to implement a machine vision solution to assessment. In this instance, we show differences in the effectiveness of existing solutions to the problem of camouflage, concealment and deception (CCD) on the battlefield. Funded by QinetiQ as part of Materials and Structures Low Observable Materials Research Programme.
Meeting abstract presented at VSS 2015