Study Of Vision Sensing System For All-Position Autonomous Welding Robot With Passing Obstacles Capability

In order to solve problems of autonomous welding in all-potion installed work piece or large non-structured equipment, it is urgent need to develop a new mobile welding robot, which has the ability to work in long distance, wide range and realize welding tasks of complicated structure such as fillet seam. Vision sensing is one of the effective ways to improve the autonomy of mobile welding robot, so a vision sensing system was developed based on the mobile welding robot platform which has the abilities of passing obstacles and all-position welding. Taking the overall coordination of the system into account, this paper fixed two network cameras onto the robot body which constituted the vehicle-eye system for monitoring the macro environment, while two CCD cameras were installed on the last arm joint which constituted hand-eye system to capture local images for intelligent welding.Calibration is the prerequisite of vision computing, this paper established the vision system calibration model according to the characteristics of mobile welding robot. Based on OpenCV, an autonomous calibration program was designed with improved hand-eye calibration algorithm and full range vehicle-eye calibration algorithm. For vehicle-eye system, obstacles were detected by a hierarchical method based on Gaussion pyramid, geometrical information was reconstructed through image process, including pre-processing, dynamic threshold segmentation, corner detection, and 3D computing of feature points. For hand-eye system, autonomous guidance was realized by two steps, which involved in the location and 3D computing of initial seam point in local region. Space complicated seam was reconstructed with discrete points'3D coordinates by SAD matching points recognition algorithm based on binocular correction model, spatial information was acquired by curve fitting.Vision sensing system was integrated onto the actual robot's hardware and software platform. Tripod head for network cameras and clamping mechanism were designed respectively, meanwhile obstacle detection and 3D information reconstruction thread, recognition of work piece for welding and guidance of initial seam point thread, spatial complicated seam reconstruction thread were wrote into overall system program. Through experimental verification, maximum errors for obstacle's distance, length, height, width were 3.08, 1.72, 2.09, 2.64mm, which could meet the requirement of passing obstacles, maximum errors for initial seam location in X, Y, Z directions were 0.81, 0.97, 1.63mm, which could satisfy the guidance requirements, maximum errors for spatial seam reconstruction in Y, Z directions were 0.15, 0.24mm. However, autonomous welding experiment integrated vision sensing capabilities showed that the combination of functional modules resulted larger errors, so further research should be carried out to study on error assessment methods and solutions to improve precision.