A Glass Detection Method Based on Multi-sensor Data Fusion in Simultaneous Localization and Mapping

Detecting glass is an important component for a mobile robot to build map of its surrounding environment, since many modern building designs feature glass panels as one of the key interior fitting elements. Aiming at the problems of glass detection in simultaneous localization and mapping (SLAM), a glass detection method based on multi-sensor data fusion, namely GD-SLAM, is proposed. Firstly, the laser point cloud with sudden intensity changes is stored in a container, and record its time, coordinates and other information. Secondly, through coordinate transformation, the coordinates of the mutation point are converted into the camera coordinate system, and the depth map within this range is obtained. Due to the use of the kinect2 camera, speckles will appear in the depth map with glass positions. Spot to confirm the existence of glass twice to improve the detection accuracy. Finally, when the camera module detects the presence of glass, the mutation point cloud recorded in this range is added to the key frame to update the point cloud. In order to confirm the performances of the method, a series of experiments in different environments are conducted. The experimental results show that, the GD-SLAM algorithm detects the glass in different environment without significantly increasing of calculation, leading to improving the accuracy of the mapping.