2017 年 136 巻 p. 144-151
The aim of this paper as a pre-stage to develop a system that can reduce the burden of ship operators classify obstacles from voyage environment images using neural network and HOG (Histogram of Oriented Gradient). Make a comparison and study with results of the obstacle classification by deep learning. We propose the obstacle classification method using the time series information, carry out the evaluation and study to classify obstacles from a river navigation video. The results are summarized as follows: (1)It was automatic classification for voyage bridge images into nine categories. (2)We conducted image classification experiments on captured images from a ship. We were obtained good correct answer rate and specific rate in the case of using HOG and neural network. (3)We conducted obstacle classification experiments using a river navigation video. We were obtained the correct answer rate of about 50 percent. (4)When the case of using time series information was obtained a good correct answer rate than not using time series information.