LRFデータの画像化処理とディープ・ラーニングを用いた周辺環境認識

抄録

This paper presents an object recognition system using deep learning for a mobile robot with LRF (Laser Range Finder). First, the quadratic distance data obtained from the LRF is flattened and classified by point groups using the developed algorithm. Next, the point groups were imaged one by one from the obtained point cloud information. These images were rotated according to the orientation of the robot. Finally, a CNN (Convolutional Neural Network) was used to recognize what these point cloud images represent. The LRF data was gathered by moving the mobile robot around many passengers and vehicles in an outdoor environment. And it was identified whether the point cloud images made from the data represent a human, a wall, a bicycle, a car, other objects.