Intensity-Based Identification of Reflective Surfaces for Occupancy Grid Map Modification

Abstract

The presence of mobile robots in indoor environments is becoming increasingly significant. While there are algorithms such as SLAM that allow robots to navigate these environments, they do not take into account structural elements of modern buildings such as large glass areas. Due to the use of lightbased sensors such as laser scans, these elements are only visible at few incident angles, so they are seen as noise by SLAM algorithms. In this work, reflective surfaces are detected by analyzing laser scan intensity values in order to build a reflection-aware map for safe indoor robot navigation. The main novelty of the proposed method is the usage of a single sensor to gather environment information and a single parameter to determine where reflective zones are. Experiments are carried out in multiple complex real scenarios, quantitatively testing how well the method performs against state of the art works and against factors such as changes in robot orientation, a large number of reflective areas or changes in illumination.

Real-Time Intensity-Based Mirror Detection for Optimizing Point Cloud Data and Occupancy Grid Map

“…In this section, four environments with mirrors are built and the performance of our method is compared with the method proposed by Mora [23] and Li [15] in these environments.In the demonstration of performance, the accuracy is the ratio of the length of the mirror calculated by the algorithm to the actual length of the mirror.…”

“…To validate the performance of the method, we constructed three complex environments and compared it with those proposed by Mora [23] and Li [15]. The experimental results show that the proposed method improves recognition accuracy and real-time performance.…”

“…Mora et al [23] experimentally obtained the intensity values of laser light at different incidence angles for three materials: metal, glass, and marble, acquired the intensity information, fitted the specular equation while the robot was moving, and modified the laser data while modifying the occupancy grid map. Although the method achieved good results in modifying the occupancy grid map, mirror detection and optimization were performed offline, and the entire mirror region could not be directly acquired during the robot motion.…”

Leveraging 3-D Data for Whole Object Shape and Reflection Aware 2-D Map Building

“…Even though these methods prove a better performance with respect to solely mapping with laser data, they have issues like a limited field of view or only detecting range differences. Other works focus on 2D data where reflectivity peaks are analyzed [9], [20]. For this to occur, sensor light beams must strike perpendicularly on the surface.…”

“…This is a continuation of previous work performed in our research group. 3D information has been previously used to generate representations at various heights [8] and glass has been detected by 2D sensors as a function of reflectivity [9]. With this new proposal, both objectives are merged and a robust occupancy grid map for robot navigation is achieved.…”