Method for recognizing predetermined object in image in e.g. robotic field, involves selecting sub-classifiers in phase by selection process and adding sub-classifiers to cascade-classifier, where cost-benefit function is used in process

摘要

The method involves using a cascade-classifier consisting of different sub-classifiers (1-3), where the sub-classifiers are formed as simple binary classifiers. A training phase and a recognition phase are carried out. The sub-classifiers are selected in the training phase by a resolution and cost-sensitive selection process and are added to the cascade-classifier. A cost-benefit function is used in the selection process, where the function serves as a break-off criterion and/or a selection criterion during the selection process. One of the sub-classifiers is formed as a linear perceptron.