Sensor-independent approach to recognition: the object-based approach

摘要

Abstract: This paper introduces a fundamentally differentapproach to recognition - the object-based approach -which is inherently knowledge-based and sensorindependent. The paper begins with a description of anobject-based recognition system, contrasting it withthe image-based approach. Next, the multilevel stage ofthe system, incorporating several sensor data sourcesis described. From these sources elements of thesituation hypothesis are generated as directed by therecognition goal. Depending on the degree ofcorrespondence between the sensor-fed elements and theobject-model-fed elements, a hypothetical element iscreated. The hypothetical element is further employedto develop evidence for the sensor-fed element throughthe inclusion of secondary sensor outputs. Thesensor-fed element is thus modeled in more detail, andfurther evidence is added to the hypothetical element.Several levels of reasoning and data integration areinvolved in this overall process; further, aself-adjusting correction mechanism is included throughthe feedback from the hypothetical element to thesensors, thus defining secondary output connections tothe sensor-fed element. Some preliminary work based onthis approach has been carried out and initial resultsshow improvements over the conventional image- basedapproach. !14