Object labeling method using uncertainty measurement

In the article is presented a new method of object labeling, based on the uncertainty measurement of a fuzzy similarity. The labeling is performed on objects detected in a scene, based on information provided by a set of different sensors. First is computed the fuzzy similarity between the detected object and a rough set of possible prototypes, followed by a measurement of the uncertainty induced by the observation. For all these results obtained from each sensor, is computed the global uncertainty corresponding to the most likely label. If we have a time stream observations about the scene and the objects or the sensor system is moving, it can be improve the labeling process by suppressing the inconsistent observations and making new labeling determinations. In this last process are used different prototypes, corresponding to different observation distances and positions. Also, from this observations performed in a time frame, it can be analyzed the uncertainty variation, determined by the switch from one prototype to another.