部位の組み合わせ学習による現場画像からの重機の種別推定システムの改良

抄録

This paper describes the way to speed up object recognition based on combinations of fragmented images for unmanned construction. Unmanned construction has been developed recently in order to recover from natural disasters such as pyroclastic flow and is good for workers on remote sites of the disaster areas because there is no risk of them when facing a secondary disaster. Construction equipment are remotely controlled with cameras which can rotate to manually track them. Therefore, an auto tracking system for construction trucks is needed in unmanned construction sites. Our prior work to realize it is recognition that types of construction equipment by using method combining Convolutional Neural Network that recognizes the object’s parts and multilayer perceptron that learns combinations of the object’s parts. However, it takes time to finish whole processing of object recognition. This paper describes considerations to speed up the processing time.