Object Recognition Using Deep Belief Nets with Spherical Signature Descriptor of 3DPoint Cloud Data for Extended Kalman Filter based Simultaneous Localization and Mapping

摘要

In previous researches on autonomous mobile robots, the data information analysis was mainly performed in terms of sensor location in order to recognize and classify surrounding objects. However, from the viewpoint of Lidar's each three-dimensional point position, not the viewpoint at the sensor position, it enables the analysis of the surrounding information at another dimension. For the purpose of object detection, we developed an Spherical Signature Descriptor (SSD) that picks up the surrounding signature of each point on an object. To learn the SSD images, we adopted Deep Belief Network (DBN) and applied it to the extended Kalman filter (EKF)-based simultaneous localization and mapping (SLAM). Experimental validation was performed using Kinect sensor data in a corridor environment.