VISAR: Vision-Based Robotic Arm System for Intelligent Industrial Inspection

Abstract

VISAR (Vision-based Intelligent System for Automated Robotic Inspection) is an intelligent inspection system that leverages a vision-equipped robotic arm to address key challenges in real-time target detection, dynamic path planning, and precise spatial localization. The system integrates a lightweight YOLOv5-ShuffleNetv2 model for efficient object recognition, combined with Canny edge detector and HSV-based color segmentation for robust target localization. Adaptive path planning is achieved using the A* algorithm, while accurate coordination between the camera and robotic arm is ensured through a nonlinear optimization-based hand-eye calibration. Implemented on the Robot Operating System (ROS) with a QT5-based user interface, the system has been validated through both Gazebo simulations and real-world tests on industrial cabinets. Results demonstrate its high efficiency, accuracy, and reliability in complex environments, showcasing strong potential for advancing automation in industrial inspection and manufacturing applications.