Image recognition of vehicles is still difficult for practical use under various actual environments. Recently
machine learning algorithm utilizing general feature amount have been often adopted. However, they utilize only a part of information obtained from images. Also, it’s difficult for human to understand the classifier, so care for individual recognition errors is hard. And so, we propose a method to design feature amounts hierarchically structuralized human’s empirical knowledge and adjust them by machine learning. Applying this method to images in actual environment, they were evaluated. As result of the experiments, 90% or more of recognition rate was achieved.
抄録全体を表示