Recognition, location, and depth estimation of objects in electrical substations

Considering the increase in using Robotic Systems with the capacity to recognize and locate objects in different areas of industry, this work presents an object recognition and location in an Electrical Energy substation (SE) developed in a virtual environment, with a strategy for depth estimation b...

Full description

Saved in:
Bibliographic Details
Main Authors: Lexinoski, Gilberto, Teixeira, Marco A. S., Lazzaretti, Andre E., Toledo, Luiz F. R. B., Lopes, Heitor S., Travassos, Gustavo, Fonseca, Mario Jambeiro
Format: Conference Proceeding
Language:English
Subjects:
Depth Estimation
Electrical Substation
Estimation
Object Location
Object Recognition
Point Cloud
Robot kinematics
Robot Operating System
Robot sensing systems
Service robots
Software
Stereo Vision
Substations
Unmanned Vehicle System
Virtual environments
YOLOv7
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Considering the increase in using Robotic Systems with the capacity to recognize and locate objects in different areas of industry, this work presents an object recognition and location in an Electrical Energy substation (SE) developed in a virtual environment, with a strategy for depth estimation by PointCloud (PC). In order to develop the environment, the CoppeliaSim software of Coppelia Robotics was used, allowing the implementation of a unit Robotic System for displacement and data acquisition in the virtual SE environment developed. Using the Robot Operating System (ROS) platform, we send commands to the UVS displacement and export the data acquired by the stereo RGB sensor to data processing in a Python environment. A YOLOv7 (You Only Look Once) model was used for object recognition and location tasks, trained to recognize and locate six types of objects present in the virtual SE environment with Bounding Boxes (BB). During the recognition and location of objects predictions, good results of the mean Average Precision (mAP) were achieved. Next, the depth estimations are performed by correlating the coordinates of the points related to the BB regions. Finally, the algorithm was implemented to predict during operation to evaluate this development's advantages and disadvantages, looking to improve the object recognition and location task during robot navigation.
ISSN:2643-685X
DOI:10.1109/LARS/SBR/WRE59448.2023.10333013